10-02-2004 12:40 AM - edited 03-02-2019 06:56 PM
Hello:
Our 3750 with 12.2(20)SE1 EMI image have process memory problem: a message appear like this:
Oct 1 23:27:23 Cat3750 180: Oct 1 23:27:22: %SYS-2-MALLOCFAIL: Memory allocation of 35856 bytes failed from 0x7B7004, alignment 8
Oct 1 23:27:23 Cat3750 181: Pool: Processor Free: 855384 Cause: Memory fragmentation
Oct 1 23:27:23 Cat3750 182: Alternate Pool: None Free: 0 Cause: No Alternate pool
Oct 1 23:27:23 Cat3750 183:
Oct 1 23:27:23 Cat3750 184: -Process= "HQM Stack Process", ipl= 0, pid= 99
Oct 1 23:27:23 Cat3750 185: -Traceback= 239FB8 23CBC8 7B7008 7A5A78 260B38 25AFE4
Oct 1 23:27:53 Cat3750 186: Oct 1 23:27:52: %SYS-2-MALLOCFAIL: Memory allocation of 35856 bytes failed from 0x7B7004, alignment 8
Oct 1 23:27:53 Cat3750 187: Pool: Processor Free: 871464 Cause: Memory fragmentation
Oct 1 23:27:53 Cat3750 188: Alternate Pool: None Free: 0 Cause: No Alternate pool
Oct 1 23:27:53 Cat3750 189:
Oct 1 23:27:53 Cat3750 190: -Process= "HQM Stack Process", ipl= 0, pid= 99
Oct 1 23:27:53 Cat3750 191: -Traceback= 239FB8 23CBC8 7B7008 7A5A78 260B38 25AFE4
Oct 1 23:28:23 Cat3750 192: Oct 1 23:28:22: %SYS-2-MALLOCFAIL: Memory allocation of 35856 bytes failed from 0x7B7004, alignment 8
Oct 1 23:28:23 Cat3750 193: Pool: Processor Free: 857168 Cause: Memory fragmentation
Oct 1 23:28:23 Cat3750 194: Alternate Pool: None Free: 0 Cause: No Alternate pool
Oct 1 23:28:23 Cat3750 195:
Oct 1 23:28:23 Cat3750 196: -Process= "HQM Stack Process", ipl= 0, pid= 99
Oct 1 23:28:23 Cat3750 197: -Traceback= 239FB8 23CBC8 7B7008 7A5A78 260B38 25AFE4
And 30 minutes later it appeared:
Oct 2 00:00:21 Cat3750 576: Oct 2 00:00:20: %FIB-3-NOMEM: Malloc Failure, disabling DCEF
Oct 2 00:01:08 Cat3750 577: Oct 2 00:01:07: %FIB-2-FIBDOWN: CEF has been disabled due to a low memory condition.
Oct 2 00:01:08 Cat3750 578: It can be re-enabled by configuring "ip cef [distributed]"
Oct 2 00:02:08 Cat3750 580: It can be re-enabled by configuring "ip cef [distributed]"
Oct 2 00:02:08 Cat3750 579: Oct 2 00:02:07: %FIB-2-FIBDOWN: CEF has been disabled due to a low memory condition.
Now I have to reload this machine to recover.
On other 3750 switch with the same configuration I saw some strange condition about the process memory uaage status: the "IP RIB Update" process hold lots of memory and didn't release:
Cat3750#sh proc mem sort
Total: 98128128, Used: 81600416, Free: 16527712
PID TTY Allocated Freed Holding Getbufs Retbufs Process
147 0 52714296 1165680 51497000 0 0 IP RIB Update
0 0 120703096 89283008 29759472 0 0 *Init*
180 0 3875849624 1793202784 4695024 402136 0 OSPF Router
183 0 1138136 16432936 1004328 0 0 BGP Router
0 0 431777456 462150776 579272 68598428 59458896 *Dead*
178 0 75798776 75267504 538192 0 0 hulc running con
......
Could anyone give me some advise?
In 3 days there are two 3750 crash for this reason and I didn't get information from CCO page.
I enable RIP/OSPF/BGP on it and have changed the SDM to "routing desktop"
Any advise is appreciate!!!
Kae Hsu
10-02-2004 02:03 AM
Answer this by myself:
It is a bug, bug ID CSCef35924
affected IOS version:
12.2(18)EW
12.2(18)EW1
12.2(18s)EW
12.2(18)SE
12.2(18)SE1
12.2(20)EW
12.2(20)SE
12.2(20)SE1
12.2(20)SE2
Workaround: reload switch
Wish Cisco can resolve this problem asap.....
03-28-2018 08:55 PM
You have the following in your configuration:
ip route 0.0.0.0 0.0.0.0 GigabitEthernet1/0/1
ip route 0.0.0.0 0.0.0.0 GigabitEthernet1/0/12 2
ip route 10.3.0.50 255.255.255.255 GigabitEthernet1/0/5
ip route 10.3.0.54 255.255.255.255 GigabitEthernet1/0/5
ip route 10.40.4.0 255.255.255.128 GigabitEthernet1/0/2
ip route 10.40.4.128 255.255.255.128 GigabitEthernet1/0/3
ip route 10.40.5.0 255.255.255.128 GigabitEthernet1/0/4
ip route 10.40.5.128 255.255.255.128 GigabitEthernet1/0/5
ip route 10.40.6.0 255.255.255.128 GigabitEthernet1/0/10
ip route 10.40.6.128 255.255.255.128 GigabitEthernet1/0/7
ip route 10.40.7.0 255.255.255.128 GigabitEthernet1/0/8
ip route 10.40.7.128 255.255.255.128 GigabitEthernet1/0/9
ip route 10.40.8.0 255.255.255.0 GigabitEthernet1/0/5
What happens here is that the connected router will proxy arp for every ip that you need to talk to across those links. This is especially bad with the default routes because you will install a /32 route for every internet route. Since the 3750 can only hold around 8k routes in the tcam (depending on the SDM template) you will start software switching and fill all the available memory with the /32 routes.
If you change the routes to be the next hop IP instead of the interface you should fix this issue.
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide