Re: Latency with 6509/msfc1 and gig servers

skhan · ‎03-19-2003

Hi,

We have a 6509 with a gigabit blade installed. We all have gig servers hanging off the 6509. When I run pings to the gig server, I get the results below where some latencys are high.

Reply from 10.1.1.195: bytes=32 time<1ms TTL=127

Reply from 10.1.1.195: bytes=32 time=428ms TTL=127

Reply from 10.1.1.195: bytes=32 time<1ms TTL=127

Reply from 10.1.1.195: bytes=32 time=942ms TTL=127

Reply from 10.1.1.195: bytes=32 time<1ms TTL=127

This mostly happens with a realport xircom rbe 100 card. I also tried the same pinging on an 6509 with msfc2 with release 7.5 with the xircom card and could not create the problem. The issue is that 6509 belongs to a company where my buddy works. I will be able to update my 6509/msfc1 OS in about a week. I am running 7.2.2 at the moment.

Just wondering, if somebody has run into this problem and what was the solution.

donewald · ‎03-21-2003

Are you certain that this problem is not due to responses from your servers being slowed by something (internal processes / etc). I've seen issues with pinging control interfaces (SC0 / etc) and getting response issues but never with transit traffic, assuming that your MLS entries are good, which I assume they are for this host.

Hope this helps,

Don

skhan · ‎03-21-2003

I believe my mls entries are fine. Here is a show mls from the Cat os

Cat-6509-41-DC> (enable) sh mls

Total packets switched = 4294967295

Total Active MLS entries = 1170

Long-duration flows aging time = 64 seconds

IP Multilayer switching aging time = 256 seconds

IP Multilayer switching fast aging time = 0 seconds, packet threshold = 0

IP Current flow mask is Full flow

Active IP MLS entries = 1170

Netflow Data Export version: 7

Netflow Data Export enabled

Netflow Data Export configured for port 2058 on host 10.1.10.85

Total packets exported = 1978441

Bridged flow statistics is disabled on vlan(s) 1-2,4,8,10,12,14,232,234-236,238,

240,242,244,246,248,250,900,999.

IP MSFC ID Module XTAG MAC Vlans

--------------- ------ ---- ----------------- ----------------

10.1.0.1 15 1 00-d0-bc-ee-e2-70 2,4,8,10,12,

999,14

IPX Multilayer switching aging time = 256 seconds

IPX flow mask is Destination flow

IPX max hop is 255

Active IPX MLS entries = 0

IPX MSFC ID Module XTAG MAC Vlans

--------------- ------ ---- ----------------- ----------------

10.1.0.1 15 1 - -

What would be your recommended mls settings.

Cat-6509-41-DC> (enable)

Other cards e.g. Intel in a desktop don't seem to have pinging issues to the server. They all come back with 1 ms response. The main problem is with the Xircom cards RBE 100. Secondly, the gig servers are not running anything at the moment. A new deployment. It is very strange. Users end up blaming that the network is slow. I have also plugged the laptop with a xircom card directly into the 6509 where the gig server resides and can still create the problem. I have isolated them completely i.e. on the 4006 and a 3524 with nothing else connected, just the laptop and the gig server and can still create the problem. I have updated the Intel gig card with the latest drivers and can still create the problem.

I am sure someone else must have a setting like mine i.e xircom cards in a laptop / 6509 and Gig servers. Wondering if they have these issues.

Don any insight to this will really be appreciated. Thanks for replying.

donewald · ‎03-21-2003

I looked for SW defects that might be causing this in your running CatOS and I see nothing that could be causing this. What type GIG module are you using that your GIg attached servers are connected? Also, have you ensured that your port counters see no errors or other issues that might show a configuration mismatch issue between the Cat and your servers?

Hope this helps,

Don

skhan · ‎03-21-2003

The card I am using is an Intel pro 1000 /xf. I have issues with both fibre and copper cards. I have no errors when I do show port on the 6509. I am using the latest drivers from Intel. The NIC cards negotiate at a 1000 /full auto.

I did find one tip that say to reduce the TCP window size, if your responses are slow from the server. This is a tcp window size for the gig card on the server. I will try that out, could or could not be related to the ping latencys.

I have some other laptop nic cards i.e. 3com, I will try as well. The fact that we have a huge xircom nic card deployment, it would get really expensive to swap, if the 3com responses better as it is a new card, newer firmware and drivers. Secondly, if I set my card to 10/half. Works great. Tried other combinations and they all have issues with ping latency.

Thanks.

donewald · ‎03-21-2003

I see nothing in CatOS that would be causing this on the Catalyst that is unresolved with regards to the GIG modules (both fiber and copper) in your 7.2.2 code.

Ping replies should not be effected by your TCP window size settings since ICMP is placed directly in IP with no transport.

It is my feeling that your issue is directly the cause of the server or NIC you are using.

Regards,

Don

skhan · ‎03-21-2003

Thanks.

I have built another server with an intel nic and still have the same problem with the Xircom cards. I will try a different nic in my laptop and try it out.

Will keep u posted. Incase you come across something please let me know.

cheers