Latency High when ping to SVI on 3850 Stack & 4500X VSS

sachin.sg · ‎02-10-2014

Dear All

We have Core->Distribution Setup ...

At Core Level We have 4500x in VSS mode and at 3850 Stack acting as Distribution Switch for Remote Building , Connectivity between Core & Distribution is 20G etherchannel ( 10G + 10G)..

We have Server Farm Vlans created on Core and User Vlans created on Cisco 3850 Distribution . To avoid STP Loops we have L3 connectivity between Core & Distribution.

We have Issue on Latency when we ping any Vlans SVI ( Gateway ) either at Core or Distribution or vice-versa latency is greater than 1ms ( 2ms or 10ms or in between ) , but if any Server from Core to any User from Distribution or within Distribution or Core itself latency is less than 1ms ,which is ok..

So is this behaviour common for new Cisco 4500X & Cisco 3850 Switches where icmp for any SVI has lower pirority , or do we need to configure something for getting latency less than 1ms....

IOS Version for 4500X :

Cisco IOS Software, IOS-XE Software, Catalyst 4500 L3 Switch Software (cat4500e-UNIVERSAL-M), Version 03.04.00.SG RELEASE SOFTWARE (fc3)

Technical Support: http://www.cisco.com/techsupport

Compiled Wed 05-Dec-12 05:58 by prod_rel_team

License Information for 'WS-C4500X-16'

License Level: entservices Type: Permanent

Next reboot license Level: entservices

IOS Version for 3850:

Package: Base, version: 03.02.02.SE, status: active

File: cat3k_caa-base.SPA.03.02.02.SE.pkg, on: Switch1

Built: Fri Jun 14 19:31:15 PDT 2013, by: udonthi

Please revert with comments , as we have seen that in 3750 if we ping SVI latency is less 1ms and Customer is saying same to us. So whats point in going for Cisco 3850 & Cisco 4500X

Jon Marshall · ‎02-10-2014

Please revert with comments , as we have seen that in 3750 if we ping SVI latency is less 1ms and Customer is saying same to us. So whats point in going for Cisco 3850 & Cisco 4500X

The point is that you don't spend a lot of time in production networks pinging the SVI itself

There are far too many points to cover in terms of benefits of 3850/4500 over 3750 but in terms of your issue when an ICMP packet is sent to to an IP address on the switch itself the ICMP is handled in software ie. it is processed by the main CPU. If you ping through a switch ie. from a client to server for example then the ICMP is hardware switched.

So using ping to the switch IPs is not necessarily a good indicator of how well the switch is performing and should not be used in any real throughput latency tests. Why the 3750 has a lower repsonse time could be that it is less busy maybe but it doesn't really matter unless the new switches are running at a high CPU rate.

If you are seeing poorer response times between end devices then that may well be an issue but it sounds like you aren't. You need to explain to the customer that what you are seeing is due to a specific way a hardware switch handles ICMP directed to one of it's own IP addresses rather than anything to do with actual hardware switch throughput.

Jon

Joseph W. Doherty · ‎02-10-2014

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Just to add to what Jon has already described, Cisco devices generally give low priority responding to pings, as pings were really more intended to indicate a device is "alive" then provide highly accurate timed responses. (NB: on some devices, Cisco supports a special ping responder [see SLA] which will note when the ping was received and try to account for device delay to provide a more accurate ping time response.)

sachin.sg · ‎02-10-2014

hi Jon

Thanks for reply ...

I have seen the ping from some client machines, which is around 2ms or 3ms for other machinces , while some machince gives less 1ms for same ...

So what can be the reason

Please provide your comments

Joseph W. Doherty · ‎02-13-2014

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

There are multiple possible reasons. Hosts aren't required to treat pings with utmost priority, so some of the variability could be just how busy the host is or there could also be congestion along the path, to name two possible issues.

Again, "basic" ping is really more to determine a host is "alive", not for highly accurate latency measurements. However, if you enable IP timestamp, you might get more accurate RTT latency (you may need to calculate the deltas, yourself).

sbhadrav@cisco.com · ‎05-08-2018

We see the 3750X stack as a single point of failure, it is a stack. The issue I would see would be downlink redundancy, as you only have 10 10g ports and you noted eleven access edge stacks and a router.

In theory, each 3750X pair of 10g ports should be able to operate at line rate. The potential bottleneck might be the StackRing of 64 Gbps. Although if just being used for the 16 10g ports, the oversubscription (5:1) isn't likely to be too excessive.

If the customer is willing to use 3850s on the edge, why didn't they consider them for the core too, as the 3850s support 4 10g per switch and there's much more ring bandwidth. (Of course the 3850s are [currently] limited to a stack of 4, but they could provide the same number of 10g ports as a stack of 8 3750X.)

However, whether using 3750Xs or 3850s, both seem an expensive way to acquire 10g ports. I.e. buy a whole switch to obtain 2 or 4 10g ports? The 4500X (or 4503) pair (of single 4507R with dual everything, but chassis) appears to be better choice for core.

You might also reconsider whether all the edge stacks need to be 3850s. Perhaps something in the new 2950X series, including the XRs, might be used for some edge stacks. If so, what you might save there might be redirected toward the core.

Could you please let me know if this question was related to a specific deal, could you also provide us with the Deal ID?