72700 - strsclnt extremely slow on Unix/Solaris (vs NT and Linux)

Status: RESOLVED WONTFIX

(NSS :: Tools, defect, P3)

Description

[Sonja Mirtitsch]

the following test was run on different machines:
a number of stressclients completed 10 connections each with the server on the
same host, no client auth. All tests took thge same time to complete:

kentuckyderby (Ultra 5/10 Ultrasparc-IIi 440MHz, Solaris 2.8 64bit, running 32
bit OPT build)
6 stressclients

sonjaNT (WinNT, 2 Processor ~700MHz, running the OPT NT build)
40 stressclients

box (Linux , 2 Processor ~700MHz)
200 stressclients

hbombaix (AIX) and charm (HP-UX) showed similar results as kentuckyderby in a
comparable test

Comment 1

[Sonja Mirtitsch]

Attachment: kentuckyderby TCP kernel info

Comment 2

[Sonja Mirtitsch]

I ran idsktune on my Solaris box and got a lot of warnings (attached). I assume
that tuning some of the parameters might increase performance, and kentuckyderby
was one of the main machines that I tested on.
More idsktune for Solaris, HP-UX, Red Hat Linux and AIX in /u/mharmsen/bin, and
http://states.red.iplanet.com/product/idsrk/idsktune/

Comment 3

[Nelson Bolyard (seldom reads bugmail)]

This bug needs a Solaris guru, which I am not.

Comment 4

[Wan-Teh Chang]

This bug should be marked WONTFIX.

This bug makes the observation that in the same amount
of time, we can run much more strsclnt processes on
dual-processor 700MHz Pentium III PCs running Linux and NT
than on a uniprocessor 440MHz UltraSPARC II running Solaris.
With twice the number of CPUs at 1.5+ times the clock
speed running a different operating system, such a discrepancy
does not surprise me.

This bug report does not state a goal.  I can make several
additional observations with the data provided (for example,
Linux can run five times the work load of NT on identical
hardware, Linux has the best price/performance ratio, etc.),
but I don't know what I am supposed to do.  Are we suspecting
that we have a performance problem in our Unix code (Solaris
in particular)?

Comment 5

[Sonja Mirtitsch]

When we ran the distributed stress test Nelson asked me to file this bug. I
would ask for some Sun people look at this issue to see if it is a problem or not.

Comment 6

[Sonja Mirtitsch]

assigning to myself before having it marked wontfix

Comment 7

[Asa Dotzler [:asa]]

why is this bug marked netscape confidential?

Comment 8

[Sonja Mirtitsch]

I don't know, but since it is not important enough for the NSS developers to
look at it it is most certainly a mistake.
Changing to everyone can see it.

Comment 9

[Sonja Mirtitsch]

marking wontfix, following Wan-Tehs recomendation, since there are no copmlaints
from the servers, and nobody with enough time and Solaris experience to really
debug this