From: Thomas Gelf (vserver_at_gelf.net)
Date: Mon 01 Mar 2004 - 07:35:16 GMT

Am Sam, den 28.02.2004 schrieb Alex Lyashkov um 09:21:
> Some question for you:
> what are maximal packet rate in you uml/tap devices ?
> what are maximal bandwidth ? what are system/user CPU usage at this time

tun/tap devices have never given me a reason to do special benchmarking
as it always worked without problems. 5 minutes ago I downloaded a 30mb
file from a real server to a uml server, both machines running somewhere
in germany in different data centers (3 hops between them). it took me
less than 3 seconds, wget said 9.59M/s. plus the overhead of http, tcp,
ip,... - we are very close to the maximum datarate possible with fast
ethernet and this over different network segments.

I've seen no significant increase of cpu usage. 3 seconds are a little
bit short to judge this, but I do not like to pay traffic for some
gigabytes to test the performance of a already well-running system.

> ?
> Using tun/tap devices need minimal a 2-4 context switches between kernel
> and user space. It`s very expensive operation in kernel and need many
> system time.

hmmm... I've no idea if you're right, but if so - it doesn't matter (to
me). how many of your customers, wich I believe to be webhosting
companies, have an' avarage network usage of more than 10 megaBYTES per
second? paying (using an' avarage european traffic price of 1 euro per
gigabyte) something like 25,000.- euros a month for network traffic???

has someone experiences with bridged tun/tap-devices on gigabit ethernet
interfaces?

cheers,
thomas

nb: I did this "test" with a p3 (uptime 800+ days :o) and a (on year
old) p4, the p4 running the umls on cheap hardware, 2,4GHz, ide and
a onboard nic.

-- 
Thomas Gelf <vserver_at_gelf.net>
_______________________________________________
Vserver mailing list
Vserver_at_list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver