On 17/10/2010 19:32, Stephan M. Mackenzie wrote: > For example, testing from Toronto, the fast servers run at about > 5.5M/s, the slow servers start at 500K/s - 1.1MB's then settle down to > as low as half that speed.
That looks like packet loss. > At peak load the router is pushing 550mb/s, cpu is at 2% and memory at > 11% ( memory has been upgraded to 1gb/1gb ) incidentally, upgrading the memory on a sup720 is almost certainly a complete waste of time and money unless you also upgrade the tcam (from 3b to 3bxl) - in which case you get the extra memory as part of the upgrade package. As you're using ws-x6148 cards and you're experiencing packet loss, you need to be aware that these line cards are desktop aggregation line cards and are _not_ intended for server farms. Please see the following URL: > http://www.cisco.com/en/US/products/hw/switches/ps700/products_tech_note09186a00801751d7.shtml#ASIC This means that internally, this line card acts a little like 6 gigabit ethernet hubs, each connected to the c6500 backplane with a 1Gb uplink. It is very easy to run into packet loss problems on this configuration. You didn't indicate in your email which ports these servers were connected up to, but assuming that they are connected into the ws-x6148, you can find out if your box is experiencing backplane drops by executing the following command for each port: #show counters interface gigabitEthernet x/y | i qos3Outlost If this shows that you're seeing packet loss, then either you need to rearrange the port configuration on your box so that none of the port groups (1-8, 9-16, 17-24, 25-32, 33-40, 41-48) bursts above 1G. Alternatively, use a ws-x6748 line card instead. Terrible performance when using ws-x6148 line cards is a FAQ. Please see the mailing list archives for more information. Nick _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
