On Thu, 2005-10-27 at 23:32 -0700, Peter Zaitsev wrote:
> Hi,
> 
> As I mentioned I'm trying to deploy  pfsense for colocation
> envinronment. 

Small followup,

Even  "agressive mode" does not seems to keep up with traffic well. 

In apache benchmark it works with concurrency=1  but fails with 30 for
example:

Completed 10000 requests
Completed 20000 requests
apr_poll: The timeout specified has expired (70007)
Total of 28227 requests completed


This behavior is repeatable - I tried rebooting the box etc.

You can also look at VMSTAT at  concurrency 30 vs 1


Concurrency 30:

 2 3 0   43064 460496  114   0   0   0  22   0   0  693  131 543  1 28
71
 procs      memory      page                   disk   faults      cpu
 r b w     avm    fre  flt  re  pi  po  fr  sr ad2   in   sy  cs us sy
id
 4 4 0   47384 458212  567   0   0   0 439   0  12 1426 2120 541  1 98
0
 1 2 0   36804 462320  436   0   0   0 641   0   9 1342  611 642  3 96
1
 0 3 0   36804 462320  653   0   0   0 636   0  19  388 2217 423  2 22
76
 0 3 0   36804 462320  638   0   0   0 620   0  19  149 2203 401  1  3
96
 1 3 0   42760 460900   94   0   0   0  22   0   0  140  116 329  0  0
99



Concurrency 1:
 1 3 0   36804 462324  649   0   0   0 633   0  19 5668 2213 10362  1 38
61
 0 3 0   36784 462324  638   0   0   0 622   0  19 6401 2204 11948  1 35
63
 0 3 0   36784 462324  638   0   0   0 619   0  19 6120 2204 11277  2 42
56
 0 3 0   36784 462324  638   0   0   0 621   0  19 5843 2203 10882  2 40
58
 2 4 0   42492 460368  526   0   0   0 415   0  12 5752  612 10613  1 36
62
 0 3 0   36784 462324  112   0   0   0 204   0   7 6393 1618 11973  0 35
65
 0 3 0   36784 462320  638   0   0   0 623   0  19 4136 2204 7715  2 27
71
 0 3 0   36784 462320  638   0   0   0 621   0  19 6074 2205 11351  2 37
61
 0 3 0   36784 462320  638   0   0   0 623   0  19 6002 2203 10880  2 36
62
 0 3 0   36804 462320  638   0   0   0 621   0  19 5485 2204 10421  2 28
70
 0 3 0   36784 462320  638   0   0   0 620   0  19 4781 2204 8802  2 29
69
 0 3 0   36804 462320 1131   0   0   0 1119   0  21 5475 2825 10282  5
38 57
 0 3 0   37012 462264  655   0   0   0 635   0  19 5775 2229 10576  1 34
66


As you see at concurrency 30 cpu usage falls significantly and  context
switches become very low....   For few first seconds CPU actually does
spike to 100%  and this is why we have some 20.000 requests completed
and when it dies...

Very interesting however even with high CPU usage number of context
switches are no where near concurrency=1 





---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to