03-29-2013 07:30 PM - edited 03-01-2019 04:54 PM
The Catalyst 4500 series switches, which includes the Catalyst 4948 switches, has a sophisticated packet-handling methodology for CPU-bound traffic. A commonly perceived problem is high CPU utilization on these switches. This document provides details about the CPU packet-handling architecture and shows you how to identify the causes of high CPU utilization on these switches. The document also lists some common network or configuration scenarios that cause high CPU utilization on the Catalyst 4500 series.
Note: If you run Catalyst OS (CatOS)-based Catalyst 4500/4000 series switches, refer to the document CPU Utilization on Catalyst 4500/4000, 2948G, 2980G, and 4912G Switches That Run CatOS Software.
There are no specific requirements for this document.
The information in this document is based on these software and hardware versions:
Note: This document applies only to Cisco IOS® Software-based switches and not CatOS-based switches.The information in this document was created from the devices in a specific lab environment. All of the devices used in this document started with a cleared (default) configuration. If your network is live, make sure that you understand the potential impact of any command.
Refer to Cisco Technical Tips Conventions for more information on document conventions.
Before you look at the CPU packet-handling architecture and troubleshoot high CPU utilization, you must understand the different ways in which hardware-based forwarding switches and Cisco IOS Software-based routers use the CPU. The common misconception is that high CPU utilization indicates the depletion of resources on a device and the threat of a crash. A capacity issue is one of the symptoms of high CPU utilization on Cisco IOS routers. However, a capacity issue is almost never a symptom of high CPU utilization with hardware-based forwarding switches like the Catalyst 4500. The Catalyst 4500 is designed to forward packets in the hardware application-specific integrated circuit (ASIC) and reach traffic-forwarding speeds of up to 102 million packets per second (Mpps).The Catalyst 4500 CPU performs these functions:
According to this list, high CPU utilization can result from the receipt or process of packets by the CPU. Some of the packets that are sent for process can be essential for the network operation. An example of these essential packets are bridge protocol data unit (BPDUs) for spanning-tree topology configurations. However, other packets can be software-forwarded data traffic. These scenarios require the switching ASICs to send packets to the CPU for processing:
The Catalyst 4500 has an in-built quality of service (QoS) mechanism in order to differentiate between types of traffic that are destined to the CPU. The mechanism makes the differentiation on the basis of the Layer 2 (L2)/Layer 3 (L3)/ Layer 4 (L4) packet information. The Supervisor packet Engine has 16 queues in order to handle various types of packets or events. Figure 1 shows these queues. Table 1 lists the queues and the packet types that queue in each. The 16 queues allow the Catalyst 4500 to queue the packets on the basis of the packet type or priority.Figure 1 – Catalyst 4500 Uses Multiple CPU QueuesTable 1 – Catalyst 4500 Queue Description
Queue Number | Queue Name | Packets Queued |
---|---|---|
0 | Esmp | ESMP1 packets (internal management packets) for the line card ASICs or other component management |
1 | Control | L2 control plane packets, such as STP, CDP, PAgP, LACP2, or UDLD3 |
2 | Host Learning | Frames with unknown source MAC addresses that are copied to the CPU in order to build the L2 forwarding table |
3, 4, 5 | L3 Fwd Highest, L3 Fwd High/Medium,L3 Fwd Low | Packets that must be forwarded in software, such as GRE4 tunnels If the ARP5 is unresolved for the destination IP address, packets are sent to this queue. |
6, 7, 8 | L2 Fwd Highest, L2 Fwd High/Medium,L2 Fwd Low | Packets that are forwarded as a result of bridging
|
9, 10 | L3 Rx High, L3 Rx Low | L3 control plane traffic, for example, routing protocols, that is destined for CPU IP addresses Examples include Telnet, SNMP, and SSH8. |
11 | RPF Failure | Multicast packets that failed the RPF9 check |
12 | ACL fwd(snooping) | Packets that are processed by the DHCP10 snooping, dynamic ARP inspection, or IGMP11 snooping features |
13 | ACL log, unreach | Packets that hit an ACE12 with the log keyword or packets that were dropped due to a deny in an output ACL or the lack of a route to the destination These packets require the generation of ICMP unreachable messages. |
14 | ACL sw processing | Packets that are punted to the CPU due to a lack of additional ACL hardware resources, such as TCAM13, for security ACL |
15 | MTU Fail/Invalid | Packets that need to be fragmented because the output interface MTU size is smaller than the size of the packet |
Use the show platform cpu packet statistics command for all troubleshooting.
Switch#show platform cpu packet statistics all!--- Output suppressed.Total packet queues 16 Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- ----...
The Catalyst 4500 CPU assigns weights to the various queues that Table 1 lists. The CPU assigns the weights on the basis of importance, or type, and on the basis of traffic priority, or DSCP. The CPU services the queue on the basis of the relative weights of the queue. For example, if both a control packet, such as a BPDU, and an ICMP echo request are pending, the CPU services the control packet first. An excessive amount of low-priority or less-important traffic does not starve the CPU of the ability to process or manage the system. This mechanism guarantees that the network is stable even under high utilization of the CPU. This ability of the network to remain stable is critical information that you must understand.
There is another very important implementation detail of Catalyst 4500 CPU packet handling. If the CPU has already serviced high-priority packets or processes but has more spare CPU cycles for a particular time period, the CPU services the low-priority queue packets or performs background processes of lower priority. High CPU utilization as a result of low-priority packet processing or background processes is considered normal because the CPU constantly tries to use all the time available. In this way, the CPU strives for maximum performance of the switch and network without a compromise of the stability of the switch. The Catalyst 4500 considers the CPU underutilized unless the CPU is used at 100 percent for a single time slot.
Cisco IOS Software Release 12.2(25)EWA2 and later have enhanced the CPU packet- and process-handling mechanism and accounting. Therefore, use these releases on your Catalyst 4500 deployments.
Now that you understand the Catalyst 4500 CPU packet-handling architecture and design, you may still wish to identify why your Catalyst 4500 CPU utilization is high. The Catalyst 4500 has the commands and tools that are necessary to identify the root cause of the high CPU utilization. After you identify the reason, the administrators can perform either of these actions:
Be sure to identify the reason for high CPU utilization even though corrective action is not necessary in all cases. High CPU utilization can be just a symptom of an issue in the network. A resolution of the root cause of that problem can be necessary in order to lower the CPU utilization.Figure 2 shows the troubleshooting methodology to use in order to identify the root cause of the Catalyst 4500 high CPU utilization.Figure 2 – High CPU Utilization Troubleshooting Methodology on Catalyst 4500 SwitchesThe general troubleshooting steps are:
The important first step is to know the CPU utilization of your switch for your configuration and network setup. Use the show processes cpu command in order to identify the CPU utilization on the Catalyst 4500 switch. The continual update of baseline CPU utilization can be necessary as you add more configuration to the network setup or as your network traffic pattern changes. Figure 2 indicates this requirement.
This output is from a fully loaded Catalyst 4507R. The steady-state CPU is about 32 to 38 percent, which is necessary in order to perform the management functions for this switch:
Switch#show processes cpu CPU utilization for five seconds: 38%/1%; one minute: 32%; five minutes: 32% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 0 63 0 0.00% 0.00% 0.00% 0 Chunk Manager 2 60 50074 1 0.00% 0.00% 0.00% 0 Load Meter 3 0 1 0 0.00% 0.00% 0.00% 0 Deferred Events !--- Output suppressed. 27 524 250268 2 0.00% 0.00% 0.00% 0 TTY Background 28 816 254843 3 0.00% 0.00% 0.00% 0 Per-Second Jobs 29 101100 5053 20007 0.00% 0.01% 0.00% 0 Per-minute Jobs 30 26057260 26720902 975 12.07% 11.41% 11.36% 0 Cat4k Mgmt HiPri 31 19482908 29413060 662 24.07% 19.32% 19.20% 0 Cat4k Mgmt LoPri 32 4468 162748 27 0.00% 0.00% 0.00% 0 Galios Reschedul 33 0 1 0 0.00% 0.00% 0.00% 0 IOS ACL Helper 34 0 2 0 0.00% 0.00% 0.00% 0 NAM Manager
Five-second CPU utilization is expressed as:
x%/y%
The x% represents total CPU utilization, and y% represents the CPU that is spent at the interrupt level. When you troubleshoot Catalyst 4500 switches, focus only on the total CPU utilization.
This show processes cpu output shows that there are two processes that use the CPU— Cat4k Mgmt HiPri and Cat4k Mgmt LoPri . These two processes aggregate multiple platform-specific processes which perform the essential management functions on the Catalyst 4500. These processes process control plane as well as data packets that need to be software-switched or processed.
In order to see which of the platform-specific processes use the CPU under the context of Cat4k Mgmt HiPri and Cat4k Mgmt LoPri , issue the show platform health command.
Each of the platform-specific processes has a target/expected utilization of the CPU. When that process is within the target, the CPU executes the process in the high-priority context. The show processes cpu command output counts that utilization under Cat4k Mgmt HiPri . If a process exceeds the target/expected utilization, that process runs under the low-priority context. The show processes cpu command output counts that additional utilization under Cat4k Mgmt LoPri . This Cat4k Mgmt LoPri is also used to run background and other low-priority processes, such as consistency check and reading interface counters. This mechanism allows the CPU to run high-priority processes when necessary, and the idle CPU cycles that remain are used for the low-priority processes. To exceed the target CPU utilization by a small amount, or a momentary spike in utilization, is not an indication of a problem that needs investigation.
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU Lj-poll 1.00 0.02 2 1 100 500 0 0 0 1:09 GalChassisVp-review 3.00 0.29 10 3 100 500 0 0 0 11:15 S2w-JobEventSchedule 10.00 0.32 10 7 100 500 0 0 0 10:14 Stub-JobEventSchedul 10.00 12.09 10 6 100 500 14 13 9 396:35 StatValueMan Update 1.00 0.22 1 0 100 500 0 0 0 6:28 Pim-review 0.10 0.00 1 0 100 500 0 0 0 0:22 Ebm-host-review 1.00 0.00 8 0 100 500 0 0 0 0:05 Ebm-port-review 0.10 0.00 1 0 100 500 0 0 0 0:01 Protocol-aging-revie 0.20 0.00 2 0 100 500 0 0 0 0:00 Acl-Flattener e 1.00 0.00 10 0 100 500 0 0 0 0:00 KxAclPathMan create/ 1.00 0.00 10 5 100 500 0 0 0 0:39 KxAclPathMan update 2.00 0.00 10 0 100 500 0 0 0 0:00 KxAclPathMan reprogr 1.00 0.00 2 0 100 500 0 0 0 0:00 TagMan-RecreateMtegR 1.00 0.00 10 0 100 500 0 0 0 0:00 K2CpuMan Review 30.00 10.19 30 28 100 500 14 13 9 397:11 K2AccelPacketMan: Tx 10.00 2.20 20 0 100 500 2 2 1 82:06 K2AccelPacketMan: Au 0.10 0.00 0 0 100 500 0 0 0 0:00 K2AclMan-taggedFlatA 1.00 0.00 10 0 100 500 0 0 0 0:00 K2AclCamMan stale en 1.00 0.00 10 0 100 500 0 0 0 0:00 K2AclCamMan hw stats 3.00 1.04 10 5 100 500 1 1 0 39:36 K2AclCamMan kx stats 1.00 0.00 10 5 100 500 0 0 0 13:40 K2AclCamMan Audit re 1.00 0.00 10 5 100 500 0 0 0 13:10 K2AclPolicerTableMan 1.00 0.00 10 1 100 500 0 0 0 0:38 K2L2 Address Table R 2.00 0.00 12 5 100 500 0 0 0 0:00 K2L2 New Static Addr 2.00 0.00 10 1 100 500 0 0 0 0:00 K2L2 New Multicast A 2.00 0.00 10 5 100 500 0 0 0 0:01 K2L2 Dynamic Address 2.00 0.00 10 0 100 500 0 0 0 0:00 K2L2 Vlan Table Revi 2.00 0.00 12 9 100 500 0 0 0 0:01 K2 L2 Destination Ca 2.00 0.00 10 0 100 500 0 0 0 0:00 K2PortMan Review 2.00 0.72 15 11 100 500 1 1 0 37:22 Gigaport65535 Review 0.40 0.07 4 2 100 500 0 0 0 3:38 Gigaport65535 Review 0.40 0.08 4 2 100 500 0 0 0 3:39 K2Fib cam usage revi 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib IrmFib Review 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib Vrf Default Ro 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib AdjRepop Revie 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib Vrf Unpunt Rev 2.00 0.01 15 0 100 500 0 0 0 0:23 K2Fib Consistency Ch 1.00 0.00 5 2 100 500 0 0 0 29:25 K2FibAdjMan Stats Re 2.00 0.30 10 4 100 500 0 0 0 6:21 K2FibAdjMan Host Mov 2.00 0.00 10 4 100 500 0 0 0 0:00 K2FibAdjMan Adj Chan 2.00 0.00 10 0 100 500 0 0 0 0:00 K2FibMulticast Signa 2.00 0.01 10 2 100 500 0 0 0 2:04 K2FibMulticast Entry 2.00 0.00 10 7 100 500 0 0 0 0:00 K2FibMulticast Irm M 2.00 0.00 10 7 100 500 0 0 0 0:00 K2FibFastDropMan Rev 2.00 0.00 7 0 100 500 0 0 0 0:00 K2FibPbr route map r 2.00 0.06 20 5 100 500 0 0 0 16:42 K2FibPbr flat acl pr 2.00 0.07 20 2 100 500 0 0 0 3:24 K2FibPbr consolidati 2.00 0.01 10 0 100 500 0 0 0 0:24 K2FibPerVlanPuntMan 2.00 0.00 15 4 100 500 0 0 0 0:00 K2FibFlowCache flow 2.00 0.01 10 0 100 500 0 0 0 0:23 K2FibFlowCache flow 2.00 0.00 10 0 100 500 0 0 0 0:00 K2FibFlowCache adj r 2.00 0.01 10 0 100 500 0 0 0 0:20 K2FibFlowCache flow 2.00 0.00 10 0 100 500 0 0 0 0:06 K2MetStatsMan Review 2.00 0.14 5 2 100 500 0 0 0 23:40 K2FibMulticast MET S 2.00 0.00 10 0 100 500 0 0 0 0:00 K2QosDblMan Rate DBL 2.00 0.12 7 0 100 500 0 0 0 4:52 IrmFibThrottler Thro 2.00 0.01 7 0 100 500 0 0 0 0:21 K2 VlanStatsMan Revi 2.00 1.46 15 7 100 500 2 2 1 64:44 K2 Packet Memory Dia 2.00 0.00 15 8 100 500 0 1 1 45:46 K2 L2 Aging Table Re 2.00 0.12 20 3 100 500 0 0 0 7:22 RkiosPortMan Port Re 2.00 0.73 12 7 100 500 1 1 1 52:36 Rkios Module State R 4.00 0.02 40 1 100 500 0 0 0 1:28 Rkios Online Diag Re 4.00 0.02 40 0 100 500 0 0 0 1:15 RkiosIpPbr IrmPort R 2.00 0.02 10 3 100 500 0 0 0 2:44 RkiosAclMan Review 3.00 0.06 30 0 100 500 0 0 0 2:35 MatMan Review 0.50 0.00 4 0 100 500 0 0 0 0:00 Slot 3 ILC Manager R 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 3 ILC S2wMan Re 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 4 ILC Manager R 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 4 ILC S2wMan Re 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 5 ILC Manager R 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 5 ILC S2wMan Re 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 6 ILC Manager R 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 6 ILC S2wMan Re 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 7 ILC Manager R 3.00 0.00 10 0 100 500 0 0 0 0:00 Slot 7 ILC S2wMan Re 3.00 0.00 10 0 100 500 0 0 0 0:00 EthHoleLinecardMan(1 1.66 0.04 10 0 100 500 0 0 0 1:18 EthHoleLinecardMan(2 1.66 0.02 10 0 100 500 0 0 0 1:18 EthHoleLinecardMan(6 1.66 0.17 10 6 100 500 0 0 0 6:38 ------------- %CPU Totals 212.80 35.63
The show platform health command provides a lot of information that is relevant only for a development engineer. In order to troubleshoot high CPU utilization, look for a higher number in the %CPU actual column in the output. Also, be sure to glance at the right side of that row in order to verify the CPU usage of that process in the 1 minute and 1 hour average %CPU columns. Sometimes, processes momentarily peak but do not hold the CPU for a long time. Some of the momentarily high CPU utilization happens during hardware programming or optimization of the programming. For example, a spike of CPU utilization is normal during the hardware programming of a large ACL in the TCAM.
In the show platform health command output in the section Understand the show processes cpu Command on the Catalyst 4500 Switches, the Stub-JobEventSchedul and the K2CpuMan Review processes use a higher number of CPU cycles. Table 2 provides some basic information about the common platform-specific processes that appear in the output of the show platform health command.
Table 2 – Description of the Platform-Specific Processes from the show platform health Command
Platform-Specific Process Name | Description |
---|---|
Pim-review | Chassis/line card state management |
Ebm | Ethernet bridge module, such as aging and monitoring |
Acl-Flattener /K2AclMan | ACL merging process |
KxAclPathMan - PathTagMan-Review | ACL state management and maintenance |
K2CpuMan Review | The process that performs software packet forwarding If you see high CPU utilization due to this process, investigate the packets that hit the CPU with use of the show platform cpu packet statistics command. |
K2AccelPacketMan | The driver that interacts with the packet engine in order to send packets that are destined from the CPU |
K2AclCamMan | Manages the input and output TCAM hardware for QoS and security features |
K2AclPolicerTableMan | Manages the input and output policers |
K2L2 | Represents the L2 forwarding subsystem of the Catalyst 4500 Cisco IOS Software These processes are responsible for maintenance of the various L2 tables. |
K2PortMan Review | Manages the various port-related programming functions |
K2Fib | FIB1 management |
K2FibFlowCache | PBR2 cache management |
K2FibAdjMan | FIB adjacency table management |
K2FibMulticast | Manages multicast FIB entries |
K2MetStatsMan Review | Manages MET3 statistics |
K2QosDblMan Review | Manages QoS DBL4 |
IrmFibThrottler Thro | IP routing module |
K2 L2 Aging Table Re | Manages the L2 aging function |
GalChassisVp-review | Chassis state monitoring |
S2w-JobEventSchedule | Manages the S2W5 protocols to monitor line cards state |
Stub-JobEventSchedul | Stub ASIC-based line card monitoring and maintenance |
RkiosPortMan Port Re | Port state monitoring and maintenance |
Rkios Module State R | Line card monitoring and maintenance |
EthHoleLinecardMan | Manages GBICs6 in each of the line cards |
1 FIB = Forwarding Information Base.
3 MET = Multicast Expansion Table.
4 DBL = Dynamic Buffer Limiting.
6 GBIC = Gigabit Interface Converter.
This section covers some of the common high CPU utilization problems on the Catalyst 4500 switches.
One of the common reasons for high CPU utilization is that the Catalyst 4500 CPU is busy with the process of packets for software-forwarded packets or control packets. Examples of software-forwarded packets are IPX or control packets, such as BPDUs. A small number of these packets is typically sent to the CPU. However, a consistently large number of packets can indicate a configuration error or a network event. You must identify the cause of events that lead to the forward of packets to the CPU for processing. This identification enables you to debug the high CPU utilization problems.
Some of the common reasons for high CPU utilization due to process-switched packets are:
Other reasons for the switch of packets to the CPU are:
The Catalyst 4500 supports 3000 spanning-tree port instances or active ports in the Per VLAN Spanning Tree+ (PVST+) mode. The support is on all Supervisor Engines, except the Supervisor Engine II+ and II+TS, and the Catalyst 4948. The Supervisor Engine II+ and II+TS and the Catalyst 4948 support up to 1500 port instances. If you exceed these STP-instance recommendations, the switch exhibits high CPU utilization.
Total number of STP instances = Number of access ports + Sum of all VLANs that are carried in each of the trunks
In the diagram, there are no access ports, but the three trunks carry VLANs 1 through 100:
Total number of STP instances = 0 + 100 + 100 + 100 = 300
This section reviews the commands that an administrator uses in order to narrow down the problem of high CPU utilization. If you issue the show processes cpucommand, you can see that two main processes, Cat4k Mgmt LoPri and Spanning Tree , primarily use the CPU. With only this information, you know that the spanning tree process consumes a sizable portion of the CPU cycles.
Switch#show processes cpu CPU utilization for five seconds: 74%/1%; one minute: 73%; five minutes: 50% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 4 198 20 0.00% 0.00% 0.00% 0 Chunk Manager 2 4 290 13 0.00% 0.00% 0.00% 0 Load Meter !--- Output suppressed. 25 488 33 14787 0.00% 0.02% 0.00% 0 Per-minute Jobs 26 90656 223674 405 6.79% 6.90% 7.22% 0 Cat4k Mgmt HiPri 27 158796 59219 2681 32.55% 33.80% 21.43% 0 Cat4k Mgmt LoPri 28 20 1693 11 0.00% 0.00% 0.00% 0 Galios Reschedul 29 0 1 0 0.00% 0.00% 0.00% 0 IOS ACL Helper 30 0 2 0 0.00% 0.00% 0.00% 0 NAM Manager !--- Output suppressed. 41 0 1 0 0.00% 0.00% 0.00% 0 SFF8472 42 0 2 0 0.00% 0.00% 0.00% 0 AAA Dictionary R 43 78564 20723 3791 32.63% 30.03% 17.35% 0 Spanning Tree 44 112 999 112 0.00% 0.00% 0.00% 0 DTP Protocol 45 0 147 0 0.00% 0.00% 0.00% 0 Ethchnl
In order to understand which platform-specific process consumes the CPU, issue the show platform health command. From this output, you can see that theK2CpuMan Review process, a job to handle CPU-bound packets, uses up the CPU:
Switch#show platform health%CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU !--- Output suppressed.TagMan-RecreateMtegR 1.00 0.00 10 0 100 500 0 0 0 0:00 K2CpuMan Review 30.00 37.62 30 53 100 500 41 33 1 2:12K2AccelPacketMan: Tx 10.00 4.95 20 0 100 500 5 4 0 0:36 K2AccelPacketMan: Au 0.10 0.00 0 0 100 500 0 0 0 0:00 K2AclMan-taggedFlatA 1.00 0.00 10 0 100 500 0 0 0 0:00
Issue the show platform cpu packet statistics command in order to check which CPU queue receives the CPU-bound packet. The output in this section shows that the control queue receives a lot of packets. Use the information in Table 1 and the conclusion that you drew in Step 1. You can determine that the packets that the CPU processes and the reason for the high CPU utilization is BPDU processing.
Switch#show platform cpu packet statistics!--- Output suppressed.Total packet queues 16 Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Esmp 202760 196 173 128 28 Control 388623 2121 1740 598 16Packets Dropped by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Control 17918 0 19 24 3
Issue the show spanning-tree summary command. You can check if the receipt of BPDUs is because of a high number of spanning-tree port instances. The output clearly identifies the root cause:
Switch#show spanning-tree summarySwitch is in pvst mode Root bridge for: none Extended system ID is enabled Portfast Default is disabled PortFast BPDU Guard Default is disabled Portfast BPDU Filter Default is disabled Loopguard Default is disabled EtherChannel misconfig guard is enabled UplinkFast is disabled BackboneFast is disabled Configured Pathcost method used is short !--- Output suppressed.Name Blocking Listening Learning Forwarding STP Active ---------------------- -------- --------- -------- ---------- ---------- 2994 vlans 0 0 0 5999 5999
There are a large number of VLANs with the PVST+ mode configuration. In order to resolve the issue, change the STP mode to Multiple Spanning Tree (MST). In some cases, the number of STP instances is high because a high number of VLANs are forwarded on all trunk ports. In this case, manually prune the VLANs that are not necessary from the trunk in order to drop the number of STP active ports to well below the recommended value.
Tip: Be sure that you do not configure IP phone ports as trunk ports. This is a common misconfiguration. Configure IP phone ports with a voice VLAN configuration. This configuration creates a pseudo trunk, but does not require you to manually prune the unnecessary VLANs. For more information on how to configure voice ports, refer to the Configuring Voice Interfaces software configuration guide. Non-Cisco IP phones do not support this voice VLAN or auxiliary VLAN configuration. You must manually prune the ports with non-Cisco IP phones.
Routing packets on the same interface, or traffic ingress and egress on the same L3 interface, can result in an ICMP redirect by the switch. If the switch knows that the next hop device to the ultimate destination is in the same subnet as the sending device, the switch generates ICMP redirect to the source. The redirect messages indicate to the source to send the packet directly to the next hop device. The messages indicate that the next hop device has a better route to the destination, a route of one less hop than this switch.
In the diagram in this section, PC A communicates with the web server. The default gateway of PC A points to the VLAN 100 interface IP address. However, the next hop router that enables the Catalyst 4500 to reach the destination is in the same subnet as PC A. The best path in this case is to send directly to "Router". Catalyst 4500 sends an ICMP redirect message to PC A. The message instructs PC A to send the packets destined to the web server via Router, instead of via Catalyst 4500. However, in most cases, the end devices do not respond to the ICMP redirect. The lack of response causes the Catalyst 4500 to spend a lot of CPU cycles on the generation of these ICMP redirects for all the packets that the Catalyst forwards via the same interface as the ingress packets.
Issue the show processes cpu command. You can see that two main processes, Cat4k Mgmt LoPri and IP Input , primarily use the CPU. With only this information, you know that the process of IP packets expends a sizable portion of the CPU.
Switch#show processes cpu CPU utilization for five seconds: 38%/1%; one minute: 32%; five minutes: 32% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 0 63 0 0.00% 0.00% 0.00% 0 Chunk Manager 2 60 50074 1 0.00% 0.00% 0.00% 0 Load Meter 3 0 1 0 0.00% 0.00% 0.00% 0 Deferred Events !--- Output suppressed. 27 524 250268 2 0.00% 0.00% 0.00% 0 TTY Background 28 816 254843 3 0.00% 0.00% 0.00% 0 Per-Second Jobs 29 101100 5053 20007 0.00% 0.01% 0.00% 0 Per-minute Jobs 30 26057260 26720902 975 5.81% 6.78% 5.76% 0 Cat4k Mgmt HiPri 31 19482908 29413060 662 19.64% 18.20% 20.48% 0 Cat4k Mgmt LoPri!--- Output suppressed. 35 60 902 0 0.00% 0.00% 0.00% 0 DHCP Snooping 36 504625304 645491491 781 72.40% 72.63% 73.82% 0 IP Input
The output of the show platform health command confirms the use of the CPU in order to process CPU-bound packets.
Switch#show platform health%CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU --- Output suppressed.TagMan-RecreateMtegR 1.00 0.00 10 0 100 500 0 0 0 0:00 K2CpuMan Review 330.00 19.18 150 79 25 500 20 19 18 5794:08 K2AccelPacketMan: Tx 10.00 4.95 20 0 100 500 5 4 0 0:36 K2AccelPacketMan: Au 0.10 0.00 0 0 100 500 0 0 0 0:00 K2AclMan-taggedFlatA 1.00 0.00 10 0 100 500 0 0 0 0:00
Issue the show platform cpu packet statistics command in order to check which CPU queue receives the CPU-bound packet. You can see that the L3 Fwd Lowqueue receives quite a lot of traffic.
Switch#show platform cpu packet statistics!--- Output suppressed.Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Esmp 48613268 38 39 38 39 Control 142166648 74 74 73 73 Host Learning 1845568 2 2 2 2 L3 Fwd High 17 0 0 0 0 L3 Fwd Medium 2626 0 0 0 0 L3 Fwd Low 4717094264 3841 3879 3873 3547L2 Fwd Medium 1 0 0 0 0 L3 Rx High 257147 0 0 0 0 L3 Rx Low 5325772 10 19 13 7 RPF Failure 155 0 0 0 0 ACL fwd(snooping) 65604591 53 54 54 53 ACL log, unreach 11013420 9 8 8 8
In this case, use the CPU SPAN in order to determine the traffic that hits the CPU. For information about the CPU SPAN, see the Tool 1: Monitor the CPU Traffic with SPAN—Cisco IOS Software Release 12.1(19)EW and Later section of this document. Complete an analysis of the traffic and a configuration with use of the show running-configuration command. In this case, a packet is routed through the same interface, which leads to the issue of an ICMP redirect for each packet. This root cause is one of the common reasons for high CPU utilization on the Catalyst 4500.
You may expect the sourcing device to act on the ICMP redirect that the Catalyst 4500 sends and change the next hop for the destination. However, not all devices respond to an ICMP redirect. If the device does not respond, the Catalyst 4500 must send redirects for every packet that the switch receives from the sending device. These redirects can consume a great deal of CPU resources. The solution is to disable ICMP redirect. Issue the no ip redirects command under the interfaces.
This scenario can occur when you also have configured secondary IP addresses. When you enable the secondary IP addresses, IP redirect is automatically disabled. Be sure you do not manually enable the IP redirects.
As this ICMP Redirects; Routing Packets on the Same Interface section has indicated, most end devices do not respond to ICMP redirects. Therefore, as a general practice, disable this feature.
The Catalyst 4500 supports IPX and AppleTalk routing via software-forwarding path only. With the configuration of such protocols, a higher CPU utilization is normal.
Note: The switching of IPX and AppleTalk traffic in the same VLAN does not require process switching. Only packets that need to be routed require software path forwarding.
Issue the show processes cpu command in order to check which Cisco IOS process consumes the CPU. In this command output, notice that the top process is theCat4k Mgmt LoPri :
witch#show processes cpu CPU utilization for five seconds: 87%/10%; one minute: 86%; five minutes: 87% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 4 53 75 0.00% 0.00% 0.00% 0 Chunk Manager !--- Output suppressed. 25 8008 1329154 6 0.00% 0.00% 0.00% 0 Per-Second Jobs 26 413128 38493 10732 0.00% 0.02% 0.00% 0 Per-minute Jobs 27 148288424 354390017 418 2.60% 2.42% 2.77% 0 Cat4k Mgmt HiPri 28 285796820 720618753 396 50.15% 59.72% 61.31% 0 Cat4k Mgmt LoPri
The output of the show platform health command confirms the use of the CPU in order to process CPU-bound packets.
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU !--- Output suppressed.TagMan-RecreateMtegR 1.00 0.00 10 4 100 500 0 0 0 0:00 K2CpuMan Review 30.00 27.39 30 53 100 500 42 47 42 4841:K2AccelPacketMan: Tx 10.00 8.03 20 0 100 500 21 29 26 270:4
In order to determine the type of traffic that hits the CPU, issue the show platform cpu packet statistics command.
Switch#show platform cpu packet statistics!--- Output suppressed.Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Esmp 48613268 38 39 38 39 Control 142166648 74 74 73 73 Host Learning 1845568 2 2 2 2 L3 Fwd High 17 0 0 0 0 L3 Fwd Medium 2626 0 0 0 0 L3 Fwd Low 1582414 1 1 1 1 L2 Fwd Medium 1 0 0 0 0 L2 Fwd Low 576905398 1837 1697 1938 1515L3 Rx High 257147 0 0 0 0 L3 Rx Low 5325772 10 19 13 7 RPF Failure 155 0 0 0 0 ACL fwd(snooping) 65604591 53 54 54 53 ACL log, unreach 11013420 9 8 8 8
Since the administrator has configured IPX or AppleTalk routing, identification of the root cause should be straightforward. But in order to confirm, SPAN the CPU traffic and be sure that the traffic that you see is the expected traffic. For information about the CPU SPAN, see the Tool 1: Monitor the CPU Traffic with SPAN—Cisco IOS Software Release 12.1(19)EW and Later section of this document.
In this case, the administrator must update the baseline CPU to the current value. The Catalyst 4500 CPU behaves as expected when the CPU processes software-switched packets.
The Catalyst 4500 learns the MAC addresses of various hosts, if the MAC address is not already in the MAC address table. The switching engine forwards a copy of the packet with the new MAC address to the CPU.
All the VLAN interfaces (layer 3) use the chassis base hardware address as their MAC address. As a result, there is not an entry in the MAC address table, and the packets destined to these VLAN interfaces are not sent to the CPU for processing.
If there is an excessive number of new MAC addresses for the switch to learn, high CPU utilization can result.
Issue the show processes cpu command in order to check which Cisco IOS process consumes the CPU. In this command output, notice that the top process is theCat4k Mgmt LoPri :
Switch#show processes cpuCPU utilization for five seconds: 89%/1%; one minute: 74%; five minutes: 71% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 4 53 75 0.00% 0.00% 0.00% 0 Chunk Manager !--- Output suppressed. 25 8008 1329154 6 0.00% 0.00% 0.00% 0 Per-Second Jobs 26 413128 38493 10732 0.00% 0.02% 0.00% 0 Per-minute Jobs 27 148288424 354390017 418 26.47% 10.28% 10.11% 0 Cat4k Mgmt HiPri 28 285796820 720618753 396 52.71% 56.79% 55.70% 0 Cat4k Mgmt LoPri
The output of the show platform health command confirms the use of the CPU in order to process CPU-bound packets.
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU !--- Output suppressed.TagMan-RecreateMtegR 1.00 0.00 10 4 100 500 0 0 0 0:00 K2CpuMan Review 30.00 46.88 30 47 100 500 30 29 21 265:01K2AccelPacketMan: Tx 10.00 8.03 20 0 100 500 21 29 26 270:4
In order to determine the type of traffic that hits the CPU, issue the show platform cpu packet statistics command.
Switch#show platform cpu packet statistics!--- Output suppressed.Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Esmp 48613268 38 39 38 39 Control 142166648 74 74 73 73 Host Learning 1845568 1328 1808 1393 1309L3 Fwd High 17 0 0 0 0 L3 Fwd Medium 2626 0 0 0 0 L3 Fwd Low 1582414 1 1 1 1 L2 Fwd Medium 1 0 0 0 0 L2 Fwd Low 576905398 37 7 8 5 L3 Rx High 257147 0 0 0 0 L3 Rx Low 5325772 10 19 13 7 RPF Failure 155 0 0 0 0 ACL fwd(snooping) 65604591 53 54 54 53 ACL log, unreach 11013420 9 8 8 8
The output of the show platform health command shows you that the CPU sees a lot of new MAC addresses. This situation is often the result of network topology instability. For example, if the spanning-tree topology changes, the switch generates Topology Change Notifications (TCNs). The issue of TCNs reduces the aging time to 15 seconds in PVST+ mode. MAC address entries are flushed if the addresses are not learned back within the time period. In the case of Rapid STP (RSTP) (IEEE 802.1w) or MST (IEEE 802.1s), the entries immediately age out if the TCN comes from another switch. This age out causes MAC addresses to be learned anew. This is not a major issue if the topology changes are rare. But there can be an excessive number of topology changes because of a flapping link, faulty switch, or host ports that are not enabled for PortFast. A large number of MAC table flushes and subsequent relearning can result. The next step in root cause identification is to troubleshoot the network. The switch works as expected and sends the packets to the CPU for host address learning. Identify and fix the faulty device that results in excessive TCNs.
Your network can have a lot of devices that send traffic in bursts, which causes MAC addresses to be aged out and subsequently relearned on the switch. In this case, increase the MAC address table aging time in order to provide some relief. With a longer aging time, the switches retain the device MAC addresses in the table for a longer period of time before the age out.
Caution: Make this age-out change only after careful consideration. The change can lead to a traffic black hole if you have devices in your network which are mobile.
The Catalyst 4500 programs the configured ACLs with use of the Cisco TCAM. TCAM allows for the application of the ACLs in the hardware-forwarding path. There is no impact on performance of the switch, with or without ACLs in the forwarding path. Performance is constant despite the size of the ACL because performance of the ACL lookups is at line rate. However, TCAM is a finite resource. Therefore, if you configure an excessive number of ACL entries, you exceed the TCAM capacity.Table 3 shows the number of TCAM resources available on each of the Catalyst 4500 Supervisor Engines and switches.
Table 3 – TCAM Capacity on Catalyst 4500 Supervisor Engines/Switches
Product | Feature TCAM (per Direction) | QoS TCAM (per Direction) |
---|---|---|
Supervisor Engine II+/II+TS | 8192 entries with 1024 masks | 8192 entries with 1024 masks |
Supervisor Engine III/IV/V and Catalyst 4948 | 16,384 entries with 2048 masks | 16,384 entries with 2048 masks |
Supervisor Engine V-10GE and Catalyst 4948-10GE | 16,384 entries with 16,384 masks | 16,384 entries with 16,384 masks |
This warning message appears when a TCAM overflow happens:
%C4K_HWACLMAN-4-ACLHWPROGERRREASON: (Suppressed 1times) Input(null, 12/Normal) Security: 140 - insufficient hardware TCAM masks. %C4K_HWACLMAN-4-ACLHWPROGERR: (Suppressed 4 times) Input Security: 140 - hardware TCAM limit, some packet processing wi...
Issue the show processes cpu command. You can see that the CPU utilization is high because the Cat4k Mgmt LoPri process takes up most of the CPU cycles.
Switch#show processes cpuCPU utilization for five seconds: 99%/0%; one minute: 99%; five minutes: 99% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 0 11 0 0.00% 0.00% 0.00% 0 Chunk Manager 2 9716 632814 15 0.00% 0.00% 0.00% 0 Load Meter 3 780 302 2582 0.00% 0.00% 0.00% 0 SpanTree Helper !--- Output suppressed. 23 18208 3154201 5 0.00% 0.00% 0.00% 0 TTY Background 24 37208 3942818 9 0.00% 0.00% 0.00% 0 Per-Second Jobs 25 1046448 110711 9452 0.00% 0.03% 0.00% 0 Per-minute Jobs 26 175803612 339500656 517 4.12% 4.31% 4.48% 0 Cat4k Mgmt HiPri 27 835809548 339138782 2464 86.81% 89.20% 89.76% 0 Cat4k Mgmt LoPri 28 28668 2058810 13 0.00% 0.00% 0.00% 0 Galios Reschedul
Issue the show platform health command. You can see that the K2CpuMan Review, a job to handle CPU-bound packets, uses the CPU.
Switch#show platform health%CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU Lj-poll 1.00 0.01 2 0 100 500 0 0 0 13:45 GalChassisVp-review 3.00 0.20 10 16 100 500 0 0 0 88:44 S2w-JobEventSchedule 10.00 0.57 10 7 100 500 1 0 0 404:22 Stub-JobEventSchedul 10.00 0.00 10 0 100 500 0 0 0 0:00 StatValueMan Update 1.00 0.09 1 0 100 500 0 0 0 91:33 Pim-review 0.10 0.00 1 0 100 500 0 0 0 4:46 Ebm-host-review 1.00 0.00 8 4 100 500 0 0 0 14:01 Ebm-port-review 0.10 0.00 1 0 100 500 0 0 0 0:20 Protocol-aging-revie 0.20 0.00 2 0 100 500 0 0 0 0:01 Acl-Flattener 1.00 0.00 10 5 100 500 0 0 0 0:04 KxAclPathMan create/ 1.00 0.00 10 5 100 500 0 0 0 0:21 KxAclPathMan update 2.00 0.00 10 6 100 500 0 0 0 0:05 KxAclPathMan reprogr 1.00 0.00 2 1 100 500 0 0 0 0:00 TagMan-InformMtegRev 1.00 0.00 5 0 100 500 0 0 0 0:00 TagMan-RecreateMtegR 1.00 0.00 10 14 100 500 0 0 0 0:18 K2CpuMan Review 30.00 91.31 30 92 100 500 128 119 84 13039:02K2AccelPacketMan: Tx 10.00 2.30 20 0 100 500 2 2 2 1345:30 K2AccelPacketMan: Au 0.10 0.00 0 0 100 500 0 0 0 0:00
You need to further understand which CPU queue and, therefore, what type of traffic hits the CPU queue. Issue the show platform cpu packet statistics command. You can see that the ACL sw processing queue receives a high number of packets. Therefore, TCAM overflow is the cause of this high CPU utilization issue.
Switch#show platform cpu packet statistics!--- Output suppressed.Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Control 57902635 22 16 12 3 Host Learning 464678 0 0 0 0 L3 Fwd Low 623229 0 0 0 0 L2 Fwd Low 11267182 7 4 6 1 L3 Rx High 508 0 0 0 0 L3 Rx Low 1275695 10 1 0 0 ACL fwd(snooping) 2645752 0 0 0 0 ACL log, unreach 51443268 9 4 5 5 ACL sw processing 842889240 1453 1532 1267 1179Packets Dropped by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- L2 Fwd Low 3270 0 0 0 0 ACL sw processing 12636 0 0 0 0
In Step 3, you determined the root cause in this scenario. Remove the ACL which caused the overflow or minimize the ACL to avoid overflow. Also, review theConfiguring Network Security with ACLs configuration guideline in order to optimize the ACL configuration and programming in the hardware.
The Catalyst 4500 supports logging of packets detail that hit any specific ACL entry, but excessive logging can cause high CPU utilization. Avoid the use of logkeywords, except during the traffic discovery stage. During the traffic discovery stage, you identify the traffic that flows through your network for which you have not explicitly configured ACEs. Do not use the log keyword in order to gather statistics. In Cisco IOS Software Release 12.1(13)EW and later, the log messages are rate-limited. If you use log messages in order to count the number of packets that match the ACL, the count is not accurate. Instead, use the show access-listcommand for accurate statistics. Identification of this root cause is easier because a review of the configuration or log messages can indicate the use of the ACL logging feature.
Issue the show processes cpu in order to check which Cisco IOS process consumes the CPU. In this command output, you find that the top process is the Cat4k Mgmt LoPri :
Switch#show processes cpuCPU utilization for five seconds: 99%/0%; one minute: 99%; five minutes: 99% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 0 11 0 0.00% 0.00% 0.00% 0 Chunk Manager 2 9716 632814 15 0.00% 0.00% 0.00% 0 Load Meter !--- Output suppressed. 26 175803612 339500656 517 4.12% 4.31% 4.48% 0 Cat4k Mgmt HiPri 27 835809548 339138782 2464 86.81% 89.20% 89.76% 0 Cat4k Mgmt LoPri 28 28668 2058810 13 0.00% 0.00% 0.00% 0 Galios Reschedul
Check the platform-specific process that uses the CPU. Issue the show platform health command. In the output, notice that the K2CpuMan Review process uses most of the CPU cycles. This activity indicates that the CPU is busy as it processes packets destined to it.
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU Lj-poll 1.00 0.01 2 0 100 500 0 0 0 13:45 GalChassisVp-review 3.00 0.20 10 16 100 500 0 0 0 88:44 S2w-JobEventSchedule 10.00 0.57 10 7 100 500 1 0 0 404:22 Stub-JobEventSchedul 10.00 0.00 10 0 100 500 0 0 0 0:00 StatValueMan Update 1.00 0.09 1 0 100 500 0 0 0 91:33 Pim-review 0.10 0.00 1 0 100 500 0 0 0 4:46 Ebm-host-review 1.00 0.00 8 4 100 500 0 0 0 14:01 Ebm-port-review 0.10 0.00 1 0 100 500 0 0 0 0:20 Protocol-aging-revie 0.20 0.00 2 0 100 500 0 0 0 0:01 Acl-Flattener 1.00 0.00 10 5 100 500 0 0 0 0:04 KxAclPathMan create/ 1.00 0.00 10 5 100 500 0 0 0 0:21 KxAclPathMan update 2.00 0.00 10 6 100 500 0 0 0 0:05 KxAclPathMan reprogr 1.00 0.00 2 1 100 500 0 0 0 0:00 TagMan-InformMtegRev 1.00 0.00 5 0 100 500 0 0 0 0:00 TagMan-RecreateMtegR 1.00 0.00 10 14 100 500 0 0 0 0:18 K2CpuMan Review 30.00 91.31 30 92 100 500 128 119 84 13039:02K2AccelPacketMan: Tx 10.00 2.30 20 0 100 500 2 2 2 1345:30 K2AccelPacketMan: Au 0.10 0.00 0 0 100 500 0 0 0 0:00
In order to determine the type of traffic that hits the CPU, issue the show platform cpu packet statistics command. In this command output, you can see that the receipt of packets is due to the ACL log keyword:
Switch#show platform cpu packet statistics!--- Output suppressed.Total packet queues 16 Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ----------------- -------------------- --------- --------- --------- ---------- Control 1198701435 35 35 34 35 Host Learning 874391 0 0 0 0 L3 Fwd High 428 0 0 0 0 L3 Fwd Medium 12745 0 0 0 0 L3 Fwd Low 2420401 0 0 0 0 L2 Fwd High 26855 0 0 0 0 L2 Fwd Medium 116587 0 0 0 0 L2 Fwd Low 317829151 53 41 31 31 L3 Rx High 2371 0 0 0 0 L3 Rx Low 32333361 7 1 2 0 RPF Failure 4127 0 0 0 0 ACL fwd (snooping) 107743299 4 4 4 4 ACL log, unreach 1209056404 1987 2125 2139 2089Packets Dropped by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ----------------- -------------------- --------- --------- --------- ---------- ACL log, unreach 193094788 509 362 437 394
In Step 3, you determined the root cause in this scenario. In order to prevent this problem, remove the log keyword from the ACLs. In Cisco IOS Software Release 12.1(13)EW1 and later, the packets are rate-limited so that CPU utilization does not get too high. Use the access list counters as a way to keep track of ACL hits. You can see the access list counters in the show access-list acl_id command output.
Layer 2 forwarding loops can be caused by poor implementation of Spanning Tree Protocol (STP) and various issues that can affect STP.
This section reviews the commands that an administrator uses in order to narrow down the problem of high CPU utilization. If you issue the show processes cpucommand, you can see that two main processes, Cat4k Mgmt LoPri and Spanning Tree , primarily use the CPU. With only this information, you know that the spanning tree process consumes a sizable portion of the CPU cycles.
Switch#show processes cpuCPU utilization for five seconds: 74%/1%; one minute: 73%; five minutes: 50% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 4 198 20 0.00% 0.00% 0.00% 0 Chunk Manager 2 4 290 13 0.00% 0.00% 0.00% 0 Load Meter !--- Output suppressed. 25 488 33 14787 0.00% 0.02% 0.00% 0 Per-minute Jobs 26 90656 223674 405 6.79% 6.90% 7.22% 0 Cat4k Mgmt HiPri 27 158796 59219 2681 32.55% 33.80% 21.43% 0 Cat4k Mgmt LoPri 28 20 1693 11 0.00% 0.00% 0.00% 0 Galios Reschedul 29 0 1 0 0.00% 0.00% 0.00% 0 IOS ACL Helper 30 0 2 0 0.00% 0.00% 0.00% 0 NAM Manager !--- Output suppressed. 41 0 1 0 0.00% 0.00% 0.00% 0 SFF8472 42 0 2 0 0.00% 0.00% 0.00% 0 AAA Dictionary R 43 78564 20723 3791 32.63% 30.03% 17.35% 0 Spanning Tree 44 112 999 112 0.00% 0.00% 0.00% 0 DTP Protocol 45 0 147 0 0.00% 0.00% 0.00% 0 Ethchnl
In order to understand which platform-specific process consumes the CPU, issue the show platform health command. From this output, you can see that theK2CpuMan Review process, a job to handle CPU-bound packets, uses up the CPU:
Switch#show platform health%CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU !--- Output suppressed.TagMan-RecreateMtegR 1.00 0.00 10 0 100 500 0 0 0 0:00 K2CpuMan Review 30.00 37.62 30 53 100 500 41 33 1 2:12K2AccelPacketMan: Tx 10.00 4.95 20 0 100 500 5 4 0 0:36 K2AccelPacketMan: Au 0.10 0.00 0 0 100 500 0 0 0 0:00 K2AclMan-taggedFlatA 1.00 0.00 10 0 100 500 0 0 0 0:00
Issue the show platform cpu packet statistics command in order to check which CPU queue receives the CPU-bound packet. The output in this section shows that the control queue receives a lot of packets. Use the information in Table 1 and the conclusion that you drew in Step 1. You can determine that the packets that the CPU processes and the reason for the high CPU utilization is BPDU processing.
Switch#show platform cpu packet statistics!--- Output suppressed.Total packet queues 16 Packets Received by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Esmp 202760 196 173 128 28 Control 388623 2121 1740 598 16Packets Dropped by Packet Queue Queue Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Control 17918 0 19 24 3
Generally, you can complete these steps in order to troubleshoot (depending on the situation, some steps are not be necessary):
Each of the steps are explained in detail at Troubleshooting Forwarding Loops - Troubleshooting STP on Catalyst Switches Running Cisco IOS System Software.
These are some other known causes of high CPU utilization:
The Catalyst 4500 exhibits high CPU utilization when one or more of the attached links starts to flap excessively. This situation occurs in Cisco IOS Software releases earlier than Cisco IOS Software Release 12.2(20)EWA.
Issue the show processes cpu command in order to check which Cisco IOS process consumes the CPU. In this command output, notice that the top process is theCat4k Mgmt LoPri :
Switch#show processes cpuCPU utilization for five seconds: 96%/0%; one minute: 76%; five minutes: 68% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 0 4 0 0.00% 0.00% 0.00% 0 Chunk Manager 2 9840 463370 21 0.00% 0.00% 0.00% 0 Load Meter 3 0 2 0 0.00% 0.00% 0.00% 0 SNMP Timers !--- Output suppressed. 27 232385144 530644966 437 13.98% 12.65% 12.16% 0 Cat4k Mgmt HiPri 28 564756724 156627753 3605 64.74% 60.71% 54.75% 0 Cat4k Mgmt LoPri 29 9716 1806301 5 0.00% 0.00% 0.00% 0 Galios Reschedul
The output of the show platform health command indicates that the KxAclPathMan create process uses up the CPU. This process is for internal path creation.
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU Lj-poll 1.00 0.03 2 0 100 500 0 0 0 9:49 GalChassisVp-review 3.00 1.11 10 62 100 500 0 0 0 37:39 S2w-JobEventSchedule 10.00 2.85 10 8 100 500 2 2 2 90:00 Stub-JobEventSchedul 10.00 5.27 10 9 100 500 4 4 4 186:2 Pim-review 0.10 0.00 1 0 100 500 0 0 0 2:51 Ebm-host-review 1.00 0.00 8 4 100 500 0 0 0 8:06 Ebm-port-review 0.10 0.00 1 0 100 500 0 0 0 0:14 Protocol-aging-revie 0.20 0.00 2 0 100 500 0 0 0 0:00 Acl-Flattener 1.00 0.00 10 5 100 500 0 0 0 0:00 KxAclPathMan create/ 1.00 69.11 10 5 100 500 42 53 22 715:0KxAclPathMan update 2.00 0.76 10 6 100 500 0 0 0 86:00 KxAclPathMan reprogr 1.00 0.00 2 1 100 500 0 0 0 0:00 TagMan-InformMtegRev 1.00 0.00 5 0 100 500 0 0 0 0:00 TagMan-RecreateMtegR 1.00 0.00 10 227 100 500 0 0 0 0:00 K2CpuMan Review 30.00 8.05 30 57 100 500 6 5 5 215:0 K2AccelPacketMan: Tx 10.00 6.86 20 0 100 500 5 5 4 78:42
Enable logging for link up/down messages. This logging is not enabled by default. The enablement helps you to narrow down the offending links very quickly. Issue thelogging event link-status command under all the interfaces. You can use the interface range command in order to conveniently enable on a range of interfaces, as this example shows:
Switch#configure terminalEnter configuration commands, one per line. End with CNTL/Z. Switch(config)#interface range gigabitethernet 5/1 - 48Switch(config-if-range)#logging event link-statusSwitch(config--if-range)#end
Switch#show logging!--- Output suppressed.3w5d: %LINK-3-UPDOWN: Interface GigabitEthernet5/24, changed state to down 3w5d: %LINK-3-UPDOWN: Interface GigabitEthernet5/24, changed state to up 3w5d: %LINK-3-UPDOWN: Interface GigabitEthernet5/24, changed state to down 3w5d: %LINK-3-UPDOWN: Interface GigabitEthernet5/24, changed state to up 3w5d: %LINK-3-UPDOWN: Interface GigabitEthernet5/24, changed state to down 3w5d: %LINK-3-UPDOWN: Interface GigabitEthernet5/24, changed state to up
After you have identified the faulty or flapping interface, shut down the interface in order to resolve the high CPU utilization issue. Cisco IOS Software Release 12.2(20)EWA and later have improved the Catalyst 4500 behavior for this flapping-links condition. Therefore, the impact on the CPU is not as great as before the improvement. Remember that this process is a background process. High CPU utilization because of this issue does not cause adverse effects on the Catalyst 4500 switches.
The Catalyst 4500 can show momentary spikes in the CPU utilization during a FIB table consistency check. The FIB table is the L3 forwarding table that the CEF process creates. The consistency check maintains consistency between the Cisco IOS Software FIB table and the hardware entries. This consistency ensures that packets are not misrouted. The check occurs every 2 seconds and runs as a low-priority background process. This process is normal behavior and does not interfere with other high-priority processes or packets.
The output of the show platform health command shows that K2Fib Consistency Ch consumes most of the CPU.
Note: The average CPU utilization for this process is insignificant over a minute or an hour, which confirms that the check is a short periodic review. This background process only uses the idle CPU cycles.
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU Lj-poll 1.00 0.02 2 1 100 500 0 0 0 1:09 GalChassisVp-review 3.00 0.29 10 3 100 500 0 0 0 11:15 !--- Output suppressed.K2Fib cam usage revi 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib IrmFib Review 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib Vrf Default Ro 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib AdjRepop Revie 2.00 0.00 15 0 100 500 0 0 0 0:00 K2Fib Vrf Unpunt Rev 2.00 0.01 15 0 100 500 0 0 0 0:23 K2Fib Consistency Ch 1.00 60.40 5 2 100 500 0 0 0 100:23 K2FibAdjMan Stats Re 2.00 0.30 10 4 100 500 0 0 0 6:21 K2FibAdjMan Host Mov 2.00 0.00 10 4 100 500 0 0 0 0:00 K2FibAdjMan Adj Chan 2.00 0.00 10 0 100 500 0 0 0 0:00 K2FibMulticast Signa 2.00 0.01 10 2 100 500 0 0 0 2:04
The Catalyst 4500 can display high CPU utilization in the K2FibAdjMan Host Move process. This high utilization appears in the output of the show platform healthcommand. Many MAC addresses frequently expire or are learned on new ports, which causes this high CPU utilization. The default value of mac-address-table aging-time is 5 minutes or 300 seconds. The workaround for this issue is to increase the MAC address aging time, or you can engineer the network in order to avoid the high number of MAC address moves. Cisco IOS Software Release 12.2(18)EW and later have enhanced this process behavior in order to consume less CPU. Refer to Cisco bug ID CSCed15021 (registered customers only) .
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU Lj-poll 1.00 0.02 2 1 100 500 0 0 0 1:09 GalChassisVp-review 3.00 0.29 10 3 100 500 0 0 0 11:15 S2w-JobEventSchedule 10.00 0.32 10 7 100 500 0 0 0 10:14 !--- Output suppressed.K2FibAdjMan Stats Re 2.00 0.30 10 4 100 500 0 0 0 6:21 K2FibAdjMan Host Mov 2.00 18.68 10 4 100 500 25 29 28 2134:39K2FibAdjMan Adj Chan 2.00 0.00 10 0 100 500 0 0 0 0:00 K2FibMulticast Signa 2.00 0.01 10 2 100 500 0 0 0 2:04 K2FibMulticast Entry 2.00 0.00 10 7 100 500 0 0 0 0:00
You can modify the maximum aging time of a MAC address in the global configuration mode. The command syntax is mac-address-table aging-time seconds for a router and mac-address-table aging-time seconds [vlan vlan-id] for a Catalyst Switch. For more information, refer to the Cisco IOS Switching Services Command Reference Guide.
The Catalyst 4500 can display high CPU utilization in the RkiosPortMan Port Review process in the output of the show platform health command in Cisco IOS Software Release 12.2(25)EWA and 12.2(25)EWA1. Cisco bug ID CSCeh08768 (registered customers only) causes the high utilization, which Cisco IOS Software Release 12.2(25)EWA2 resolves. This process is a background process and does not affect the stability of the Catalyst 4500 switches.
Switch#show platform health %CPU %CPU RunTimeMax Priority Average %CPU Total Target Actual Target Actual Fg Bg 5Sec Min Hour CPU Lj-poll 1.00 0.02 2 1 100 500 0 0 0 1:09 GalChassisVp-review 3.00 0.29 10 3 100 500 0 0 0 11:15 S2w-JobEventSchedule 10.00 0.32 10 7 100 500 0 0 0 10:14 !--- Output suppressed.K2 Packet Memory Dia 2.00 0.00 15 8 100 500 0 1 1 45:46 K2 L2 Aging Table Re 2.00 0.12 20 3 100 500 0 0 0 7:22 RkiosPortMan Port Re 2.00 87.92 12 7 100 500 99 99 89 1052:36Rkios Module State R 4.00 0.02 40 1 100 500 0 0 0 1:28 Rkios Online Diag Re 4.00 0.02 40 0 100 500 0 0 0 1:15
If a port is configured for both the voice VLAN option and the access VLAN option, the port acts as a multi-VLAN access port. The advantage is that only those VLANs that are configured for the voice and access VLAN options are trunked.
The VLANs that are trunked to the phone increase the number of STP instances. The switch manages the STP instances. Management of the increase in STP instances also increases the STP CPU utilization.
The trunking of all the VLANs also causes unnecessary broadcast, multicast, and unknown unicast traffic to hit the phone link.
Switch#show processes cpuCPU utilization for five seconds: 69%/0%; one minute: 72%; five minutes: 73% PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process 1 4 165 24 0.00% 0.00% 0.00% 0 Chunk Manager 2 29012 739091 39 0.00% 0.00% 0.00% 0 Load Meter 3 67080 13762 4874 0.00% 0.00% 0.00% 0 SpanTree Helper 4 0 1 0 0.00% 0.00% 0.00% 0 Deferred Events 5 0 2 0 0.00% 0.00% 0.00% 0 IpSecMibTopN 6 4980144 570766 8725 0.00% 0.09% 0.11% 0 Check heaps 26 539173952 530982442 1015 13.09% 13.05% 13.20% 0 Cat4k Mgmt HiPri 27 716335120 180543127 3967 17.61% 18.19% 18.41% 0 Cat4k Mgmt LoPri 33 1073728 61623 17424 0.00% 0.03% 0.00% 0 Per-minute Jobs 34 1366717824 231584970 5901 38.99% 38.90% 38.92% 0 Spanning Tree 35 2218424 18349158 120 0.00% 0.03% 0.02% 0 DTP Protocol 36 5160 369525 13 0.00% 0.00% 0.00% 0 Ethchnl 37 271016 2308022 117 0.00% 0.00% 0.00% 0 VLAN Manager 38 958084 3965585 241 0.00% 0.01% 0.01% 0 UDLD 39 1436 51011 28 0.00% 0.00% 0.00% 0 DHCP Snooping 40 780 61658 12 0.00% 0.00% 0.00% 0 Port-Security 41 1355308 12210934 110 0.00% 0.01% 0.00% 0 IP Input
Layer 3 control packets that are captured with RSPAN are destined to CPU rather than just the RSPAN destination interface, which causes high CPU. The L3 control packets are captured by static CAM entries with forward to CPU action. The static CAM entries are global to all VLANs. In order to avoid unnecessary CPU flooding, use the Per-VLAN Control Traffic Intercept feature, available in Cisco IOS software releases 12.2(37)SG and later.
Switch(config)# access-list hardware capture mode vlan
Static ACLs are installed at the top in input feature TCAM to capture control packets destined to well known IP multicast addresses in the 224.0.0.* range. Static ACLs are installed at boot time and appear before any user configured ACL. Static ACLs are always hit first and intercept control traffic to CPU on all VLANs.
Per-VLAN control traffic intercept feature provide selective per-VLAN path managed mode of capturing control traffic. The corresponding static CAM entries in input feature TCAM are invalidated in the new mode. Control packets are captured by feature specific ACL attached to VLANs on which snooping or routing features are enabled. There is no feature specific ACL attached to RSPAN VLAN. Therefore, all layer 3 control packets received from RSPAN VLAN are not forwarded to CPU.
As this document has shown, traffic that is destined to the CPU is one of the major causes of high CPU utilization on the Catalyst 4500. The CPU-destined traffic can be either intentional because of the configuration, or unintentional because of misconfiguration or a denial-of-service attack. The CPU has an in-built QoS mechanism to prevent any adverse network effects because of this traffic. However, identify the root cause of CPU-bound traffic and eliminate the traffic if it is undesirable.
The Catalyst 4500 allows for the monitor of the CPU-bound traffic, either ingress or egress, with the use of the standard SPAN function. The destination interface connects to a packet monitor or an administrator laptop that runs packet sniffer software. This tool helps to quickly and accurately analyze the traffic that the CPU processes. The tool provides the ability to monitor individual queues that are bound to the CPU packet engine.
Note: The switching engine has 32 queues for the CPU traffic, and the CPU packet engine has 16 queues.
Switch(config)#monitor session 1 source cpu ? both Monitor received and transmitted traffic queue SPAN source CPU queue rx Monitor received traffic only tx Monitor transmitted traffic only <cr> Switch(config)#monitor session 1 source cpu queue ? <1-32> SPAN source CPU queue numbers acl Input and output ACL [13-20] adj-same-if Packets routed to the incoming interface [7] all All queues [1-32] bridged L2/bridged packets [29-32] control-packet Layer 2 Control Packets [5] mtu-exceeded Output interface MTU exceeded [9] nfl Packets sent to CPU by netflow (unused) [8] routed L3/routed packets [21-28] rpf-failure Multicast RPF Failures [6] span SPAN to CPU (unused) [11] unknown-sa Packets with missing source address [10] Switch(config)#monitor session 1 source cpu queue all rxSwitch(config)#monitor session 1 destination interface gigabitethernet 1/3Switch(config)#end4w6d: %SYS-5-CONFIG_I: Configured from console by console Switch#show monitor session 1Session 1 --------- Type : Local Session Source Ports : RX Only : CPU Destination Ports : Gi1/3 Encapsulation : Native Ingress : Disabled Learning : Disabled
If you connect a PC that runs a sniffer program, you can quickly analyze the traffic. In the output that appears in the window in this section, you can see that the cause of the high CPU utilization is an excessive number of STP BPDUs.
Note: STP BPDUs in the CPU sniffer is normal. But if you see more than you expect, you may have exceeded the recommended limits for your Supervisor Engine. See the A High Number of Spanning-Tree Port Instances section of this document for more information.
The Catalyst 4500 provides an in-built CPU sniffer and decoder to quickly identify the traffic that hits the CPU. You can enable this facility with the debug command, as the example in this section shows. This features implements a circular buffer that can retain 1024 packets at a time. As new packets arrive, they overwrite the older packets. This feature is safe to use when you troubleshoot high CPU utilization issues.
Switch#debug platform packet all receive bufferplatform packet debugging is on Switch#show platform cpu packet bufferedTotal Received Packets Buffered: 36 ------------------------------------- Index 0: 7 days 23:6:32:37214 - RxVlan: 99, RxPort: Gi4/48Priority: Crucial, Tag: Dot1Q Tag, Event: Control Packet, Flags: 0x40, Size: 68 Eth: Src 00-0F-F7-AC-EE-4F Dst 01-00-0C-CC-CC-CD Type/Len 0x0032 Remaining data: 0: 0xAA 0xAA 0x3 0x0 0x0 0xC 0x1 0xB 0x0 0x0 10: 0x0 0x0 0x0 0x80 0x0 0x0 0x2 0x16 0x63 0x28 20: 0x62 0x0 0x0 0x0 0x0 0x80 0x0 0x0 0x2 0x16 30: 0x63 0x28 0x62 0x80 0xF0 0x0 0x0 0x14 0x0 0x2 40: 0x0 0xF 0x0 0x0 0x0 0x0 0x0 0x2 0x0 0x63 Index 1: 7 days 23:6:33:180863 - RxVlan: 1, RxPort: Gi4/48 Priority: Crucial, Tag: Dot1Q Tag, Event: Control Packet, Flags: 0x40, Size: 68 Eth: Src 00-0F-F7-AC-EE-4F Dst 01-00-0C-CC-CC-CD Type/Len 0x0032 Remaining data: 0: 0xAA 0xAA 0x3 0x0 0x0 0xC 0x1 0xB 0x0 0x0 10: 0x0 0x0 0x0 0x80 0x0 0x0 0x2 0x16 0x63 0x28 20: 0x62 0x0 0x0 0x0 0x0 0x80 0x0 0x0 0x2 0x16 30: 0x63 0x28 0x62 0x80 0xF0 0x0 0x0 0x14 0x0 0x2 40: 0x0 0xF 0x0 0x0 0x0 0x0 0x0 0x2 0x0 0x63
Note: The CPU utilization when you issue a debug command is always almost 100%. It is normal to have high CPU utilization when you issue a debug command.
Catalyst 4500 provides another useful tool to identify the top interfaces that send traffic/packets for CPU processing. This tool helps you quickly identify an errand device that sends a high number of broadcast or other denial-of-service attacks to the CPU. This feature is also safe to use when you troubleshoot high CPU utilization issues.
Switch#debug platform packet all countplatform packet debugging is on Switch#show platform cpu packet statistics!--- Output suppressed.Packets Transmitted from CPU per Output Interface Interface Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Gi4/47 1150 1 5 10 0 Gi4/48 50 1 0 0 0 Packets Received at CPU per Input InterfaceInterface Total 5 sec avg 1 min avg 5 min avg 1 hour avg ---------------------- --------------- --------- --------- --------- ---------- Gi4/47 23130 5 10 50 20 Gi4/48 50 1 0 0 0
Note: The CPU utilization when you issue a debug command is always almost 100%. It is normal to have high CPU utilization when you issue a debug command.
The Catalyst 4500 switches handle a high rate of IP version 4 (IPv4) packet forwarding in hardware. Some of the features or exceptions can cause the forward of some packets via the CPU process path. The Catalyst 4500 uses a sophisticated QoS mechanism to handle CPU-bound packets. This mechanism ensures reliability and stability of the switches and, at the same time, maximizes the CPU for the software forwarding of packets. Cisco IOS Software Release 12.2(25)EWA2 and later provide additional enhancements for packet/process handling as well as accounting. The Catalyst 4500 also has sufficient commands and powerful tools to aid in the identification of the root cause of high CPU-utilization scenarios. But, in most cases, high CPU utilization on the Catalyst 4500 is not a cause of network instability nor a cause for concern.
Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: