CSM slow start not working or not slow enough

[ecollege]

Here's the basic config:

-serverfarm A is configured with leastconns, default slow start timer (1).

-serverfarm A has 18 servers added to it.

-variable REAL_SLOW_START_ENABLE is at 3, the default.

-IOS version: 12.2(18)SXD5

-CSM OS version: 4.2(2)

Here's the problem:

-When taking a server out of service and putting it back in, the server ramps up to the average connections so fast that the server's CPU spikes and many times it fails.

-The connections go from 0 to ~500 (the average at the time) in about a minute or less.

What we've tried:

-Setting the REAL_SLOW_START_ENABLE variable to 1, 2, 5, 9, with no apparent difference in ramp up speed.

-Setting the slowstart timer for serverfarm A to 300 seconds, with no apparent difference in ramp up speed.

-We're aware of caveats CSCef76686 and CSCei37874, but changing the variable value to 2, 5, or 9 should have negated those.

Anyone have any thoughts? Thank you.

[pgolding]

Brad,

This could be CSCei26434, workaround is to use roundrobin predictor. The bug is resolved in 4.2(2.6) so if you have that or later, its something else.

Peter

[ecollege]

Actually we ran into that bug separately, and doing a "no in" and"in" is a workaround. But this issue is different, it's more dealing with how fast the servers are taking on new connections once they're put in service.

[pgolding]

Brad,

There does not appear to be any other existing bug that matches this description, so I suggest you open a case with the TAC if you have not done so already.

Peter