Question

Locked

Why are some clients slow to connect to servers right next to each other?

By gillogly ·
OK, here's the deal. We just installed 2 new win2k3 servers clustered and were migrating from an old win2k3 server all files and shares. All 3 servers are located on a cisco gig switch. Cluster servers will be considered server A and B. Old server will be server C. Some clients are really slow connecting to servers A and B. Decent speeds connecting to server C. Here are some tests with ttcpw.exe, the utility that cisco uses.

Server A ? Host 1

C>ttcpw -t -s *Host 1*
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp -> *Host 1*

ttcp-t: socket
ttcp-t: connect
ttcp-t: 16777216 bytes in 66.33 seconds = 247.01 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 33.16, calls/sec = 30.88
ttcp-t:

C>ttcpw -r -s
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp
ttcp-r: socket
ttcp-r: accept from *Host 1*
ttcp-r: 16777216 bytes in 15.11 seconds = 1084.39 KB/sec +++
ttcp-r: 2235 I/O calls, msec/call = 6.92, calls/sec = 147.93

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

Server A ? Host 2

C>ttcpw -r -s
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp
ttcp-r: socket
ttcp-r: accept from *Host 2*
ttcp-r: 16777216 bytes in 1.44 seconds = 11401.53 KB/sec +++
ttcp-r: 2775 I/O calls, msec/call = 0.53, calls/sec = 1931.11
ttcp-r:

C>ttcpw -t -s *Host 2*
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp -> *Host 2*

ttcp-t: socket
ttcp-t: connect
ttcp-t: 16777216 bytes in 1.42 seconds = 11521.80 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 0.71, calls/sec = 1440.23
ttcp-t:

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

Server C - Host 1

U>ttcpw -t -s *Host 1*
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp -> *Host 1*

ttcp-t: socket
ttcp-t: connect
ttcp-t: 16777216 bytes in 23.83 seconds = 687.59 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 11.91, calls/sec = 85.95
ttcp-t:

U>ttcpw -r -s
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp
ttcp-r: socket
ttcp-r: accept from *Host 1*
ttcp-r: 16777216 bytes in 15.14 seconds = 1082.09 KB/sec +++
ttcp-r: 2084 I/O calls, msec/call = 7.44, calls/sec = 137.64
ttcp-r:

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

Server A - Host 1

U>ttcpw -t -s *Host 1*
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp -> *Host 1*

ttcp-t: socket
ttcp-t: connect
ttcp-t: 16777216 bytes in 82.91 seconds = 197.62 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 41.45, calls/sec = 24.70
ttcp-t:

U>ttcpw -r -s
ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp
ttcp-r: socket
ttcp-r: accept from *Host 1*
ttcp-r: 16777216 bytes in 15.13 seconds = 1083.24 KB/sec +++
ttcp-r: 2256 I/O calls, msec/call = 6.87, calls/sec = 149.16
ttcp-r:

This conversation is currently closed to new comments.

4 total posts (Page 1 of 1)  
| Thread display: Collapse - | Expand +

All Answers

Collapse -

Couple of things..

by bart777 In reply to Why are some clients slow ...

First make sure that the NICs are hard coded to match the ports on the GIG switch. Sometimes the server nics will do strange things and you get slow or bursty connections.

Next if you are using NIC teaming on the servers with dual NICs you will need to go into the Cisco device and change the teaming protocol. I've seen this odd behavior before but don't know the protocal name. Cisco uses it's own teaming protcol and companies like Dell use another. Cisco chan help you with this. A quick way to test this is to break the team at the server end and see if the response comes back to normal.

Give them a whirl.

Best of luck

Collapse -

Re: Couple of things..

by gillogly In reply to Couple of things..

As far as statically assigning duplex/speed on server, no can do. Options are only auto, 10/100 half/full. I set the cisco ports to 1000 full as a test and there is no change. It appears that only a few of the clients have slow speeds receiving only from the cluster servers. The old server is still 7x or more faster in throughput.

Thanks

Collapse -

Flow Control!

by gillogly In reply to Why are some clients slow ...

We found that for some reason flow control was enabled on the NICs.

Back to Networks Forum
4 total posts (Page 1 of 1)  

Related Discussions

Related Forums