[uml-user] Long creation of threads

[Marcin <mi...@op...>]

Hi,
I'm trying to run a proprietary application written for my company under 
UML. The main program is written in Kylix and the executable is quite 
large (almost 100MB - yes I mean megabytes ;)). The application acts as 
daemon and spawn a thread to process each incoming query. The 
application makes no i/o processing -- it just receive query, computes
result and sends it back.

I noticed a very poor performance while running it under UML. The 
environment is as follows:

Host:
2xPIII 1.4GHz
1250MB RAM
2*73GB HDD SCSI3 10k RPM in software RAID1
2.6.10 with SKAS3/2.6-V8-rc2
tmpfs mounted on /tmp with size 640MB

Guest:
2.6.9bs7
460MB RAM
dedicated partition on 18GB HDD SCSI3

The host has no other services running (except SSH for access). It's a 
backup box for our main PostgreSQL server, but with no PSQL daemon 
running (just waiting for a potential failure). So it's not busy at all.
Some further investigation shows, that application performance suffer 
strongly from very long thread creation. On host machine it takes few 
(5-7) ms to spawn another thread, while under UML it is over 100ms.

The lsof shows that UML is using tmpfs for vm files, and the SKAS patch 
is compiled into host kernel.

Is there any chance to improve performance of that app (I can't force 
the vendor to decrease size of the executable).

Thanks for help,

-- 
Marcin