LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: Does SMP significantly increase LVS performance ?

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: Re: Does SMP significantly increase LVS performance ?
From: Kyle Sparger <ksparger@xxxxxxxxxxxxxxxxxxxx>
Date: Mon, 13 Mar 2000 08:43:20 -0500 (EST)
I don't know if this still holds true or not, but as I recall, at one
point, a big part of what happened with the Mindcraft benchmarks was
blamed on the fact that the kernel's TCP/IP stack is (was?) not
multithreaded.

If this is the case, I would imagine that extra CPU's would be wasted
on the director until such a time as the stack becomes threaded -- I
believe 2.3/2.4 solves this problem.

Considering that information, if I was concerned that a director I was
building was going to be CPU bound, I'd recommend spending the
extra money on a faster CPU cache -- ie, buy a PIII 600eb or better, a
K6-3, a Xeon, or even a Celeron (probably in that order), rather than on
an SMP system.  I believe that the high-speed, high associativity of the
Coppermine chips makes them ideal for a director.  I bet you could keep
all (or nearly so) of the relevant kernel code in cache 100% of the time
on these things, which means you probably won't have wait on data from RAM
to push those packets, which in turn means far more efficient use of the
CPU cycles available.

Faster, lower latency RAM (PC133 SDRAM, not RDRAM) wouldn't hurt either,
but might be overkill.

Also, what Joe said is 100% correct -- unless you have a very peculiar 
case, this is going to be I/O bound at the network level, rather than CPU
bound, so worry about that first.  

Thanks,

Kyle Sparger - Senior System Administrator
Dialtone Internet - Extremely Fast Web Systems
(954) 581-0097 - Voice (954) 581-7629 - Fax
ksparger@xxxxxxxxxxxxxxxxxxxx
http://www.dialtoneinternet.net

On Sat, 11 Mar 2000, Joseph Mack wrote:

> On Thu, 9 Mar 2000, CERBA Jacques -GRE wrote:
> 
> > Doe's somebody have any idea or data ?
> 
> it depends :-)
> 
> If you're using VS-NAT then you'll need a machine that can handle the full
> bandwidth of the expected connections. If this is T1, you won't need much
> of a machine. If it's 100Mbps you'll need more (I can saturate 100Mbps
> with a 75MHz machine). If you're running VS-DR or VS-Tun you'll need
> less horse power. Since most LVS is I/O I would suspect that SMP won't
> get you much. However if the director is doing other things too,
> then SMP might be useful
> 
> Joe
> 
> --
> Joseph Mack mack@xxxxxxxxxxx
> 
> 
> 







<Prev in Thread] Current Thread [Next in Thread>