This message was deleted.
# citrix-netscalers
Slackbot
06/24/2020, 9:10 AMThis message was deleted.
e
Eirik Vesterhus / Intility AS
06/24/2020, 9:25 AMAlso, it ran fine in block mode for several months without this problem occuring. And that was with more signatures enabled.
c
CTXMike
06/24/2020, 2:02 PMIf this ticket is still with frontline, request escalation. 3 dedicated CPUs doesn't make sense, expecially on an SDX where I'm assuming it also already has crypto assigned to it. The bandwidth doesn't make sense either as it has no effect on performance on an SDX unless you're actually hitting a rate limit.
e
Eirik Vesterhus / Intility AS
06/24/2020, 4:13 PMIt has been escalated for quite some time. Yes, it has crypto assigned to it. I have replied and voiced my concerns about the proposed solution, but knowing the case history I probably will not get an answer I am pleased with.
Eirik Vesterhus / Intility AS
07/27/2020, 11:55 AMWe do hit the platform rate limit occasionally (3-4 events a month). It is probably some of our customers doing load testing. I do not have a problem with it. The ADC should be able to handle it by not shitting on itself. To add - neither we or Citrix can prove that the tight loop is connected to the platformRateLimitPktDrop events.
Anyway - if this means the ADC can not handle a sudden increase in traffic without entering a tight loop, I would rather decrease the throughput and assigning 2 CPU cores rather than increasing and adding 3 CPU cores.
We will see how it goes. Reporting back when I have some new information.
c
CTXMike
07/27/2020, 3:06 PMThe whole point of the device is to gracefully handle traffic surges... even just two PEs can chew a lot of throughput. If you're doing something really heavy like AppFW with all the fancy knobs enabled then that might be an issue, but that's something that they should be able to determine easily through normal CPU profiling.
e
Eirik Vesterhus / Intility AS
08/07/2020, 11:12 AMI agree! Unfortunately - CPU profiling did not give us an explanation for the problems. We ended up adding another CPU core to the VPX (2 in total, dedicated) and decreasing the throughput a bit. Seems better now, but we'll see.
Eirik Vesterhus / Intility AS
08/12/2020, 1:36 PMTurns out it was because of a custom log action.
Log buffers got exhausted when logging HTTP.REQ.BODY(HTTP.REQ.CONTENT_LENGTH)
9 Views