how to test the performance of modified openssh

[Daniel Kahn Gillmor]

On Fri 2008-01-11 14:55:52 -0500, Ben Lindstrom wrote:

> In both cases I believe the test was pretty much a "dd if=/dev/urandom .. 
> | ssh '| cat /dev/null'" style testing or transferring of real data. 
> Timing the time it takes to move XX amount of data then doing the math.

Be aware that if you're concerned about a CPU bottleneck for openssh,
dd'ing from /dev/urandom is going to confound your results.  Pulling
only 5MB of data from /dev/urandom chews up over 2.5 seconds of CPU
time (inside the kernel, no less!) on a 1.2GHz P3 running Linux
2.6.22.  here's two different ways of looking at it:

[0 dkg at squeak ~]$ time dd if=/dev/urandom bs=1K count=5K > /dev/null 
5120+0 records in
5120+0 records out
5242880 bytes (5.2 MB) copied, 2.69792 seconds, 1.9 MB/s

real	0m2.723s
user	0m0.016s
sys	0m2.696s
[0 dkg at squeak ~]$ (sleep 2 && dd if=/dev/urandom bs=1K count=5K > /dev/null 2>/dev/null) & vmstat 1 10
[1] 14475
procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa
 2  0 420188   6352   4776  76720    3    2    52    32    5    2  7  1 90  2
 0  0 420188   6364   4776  76720    0    0     0     0   75  206  0  0 100  0
 1  0 420188   6364   4776  76720    0    0     0     0   75  117  0  2 98  0
 1  0 420188   6364   4776  76720    0    0     0     0  255   83  0 100  0  0
 1  0 420188   6364   4776  76720    0    0     0     0  266  111  0 100  0  0
 0  0 420188   6432   4776  76720    0    0     0     0  278  143  0 67 33  0
 0  0 420188   6452   4776  76720    0    0     0     0  109  131  0  0 100  0
 0  0 420188   6452   4776  76720    0    0     0     0   87  119  0  0 100  0
 0  0 420188   6452   4776  76720    0    0     0     0  212  213  0  1 99  0
 1  0 420188   6452   4776  76720    0    0     0     0  132  152  0  0 100  0
[1]+  Done                    ( sleep 2 && dd if=/dev/urandom bs=1K count=5K >/dev/null 2>/dev/null )
[0 dkg at squeak ~]$ 

So if ssh is contending with the kernel's PRNG and the CPU is pegged
at 100%, you'll see a throughput performance degradation as a result.

It's probably better to dd from /dev/urandom to a large (RAM-backed)
file and feed a continuous loop from the file for throughput tests to
avoid this problem.

           --dkg
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 826 bytes
Desc: not available
Url : http://lists.mindrot.org/pipermail/openssh-unix-dev/attachments/20080111/24aaaec3/attachment.bin