"Feldman, Jim" wrote:
>
> The current (probably April 2001) issue of Linux Journal has an article from
> a pair of Erikson engineers who were looking at using LVS to front end a
> bunch of diskless web servers. They only tried NAT in the article, and on a
> 500mhz pIII got about 1700-1800 connections per second before it topped out
> (tested using Web Bench). Direct connection to the servers yielded around
> 8000 connects/sec. What wasn't clear was where the bottleneck was.
They didn't try to find out either.
Would a
> Ghz class pIII double the throughput, would better NIC's help, is there a
> limitation of the Linux stack? They were going to try the DR mode, but
> hadn't by press time.
Let's assume a lead time of 3 months for publication. You are looking at an
article by research engineers who couldn't get VS-DR working in 3 months using
a precompiled kernel, with lvs-gui which sets up VS-DR for you, when they
already
had VS-NAT working.
They quote similar latency and throughput per server to what I got for a 75MHz
pentium classic real-server (my performance page on the website).
> I believe they were using the stock 6.2 RedHat
> Piranha dist. The 2.4 kernel is reputed to have a more efficient network
> stack, does anyone have a feel for the amount of perf improvement for LVS
> (understanding that Web Bench is a benchmark and nothing more)?
benchmarks are benchmarks. If you want to see what the limitations are try
http://www.linuxvirtualserver.org/Joseph.Mack/HOWTO/LVS-HOWTO_1.0-5.html#ss5.2
(it may not be terribly clear) then come up and ask again
Joe
--
Joseph Mack PhD, Senior Systems Engineer, Lockheed Martin
contractor to the National Environmental Supercomputer Center,
mailto:mack.joseph@epa.gov ph# 919-541-0007, RTP, NC, USA
>
> The current (probably April 2001) issue of Linux Journal has an article from
> a pair of Erikson engineers who were looking at using LVS to front end a
> bunch of diskless web servers. They only tried NAT in the article, and on a
> 500mhz pIII got about 1700-1800 connections per second before it topped out
> (tested using Web Bench). Direct connection to the servers yielded around
> 8000 connects/sec. What wasn't clear was where the bottleneck was.
They didn't try to find out either.
Would a
> Ghz class pIII double the throughput, would better NIC's help, is there a
> limitation of the Linux stack? They were going to try the DR mode, but
> hadn't by press time.
Let's assume a lead time of 3 months for publication. You are looking at an
article by research engineers who couldn't get VS-DR working in 3 months using
a precompiled kernel, with lvs-gui which sets up VS-DR for you, when they
already
had VS-NAT working.
They quote similar latency and throughput per server to what I got for a 75MHz
pentium classic real-server (my performance page on the website).
> I believe they were using the stock 6.2 RedHat
> Piranha dist. The 2.4 kernel is reputed to have a more efficient network
> stack, does anyone have a feel for the amount of perf improvement for LVS
> (understanding that Web Bench is a benchmark and nothing more)?
benchmarks are benchmarks. If you want to see what the limitations are try
http://www.linuxvirtualserver.org/Joseph.Mack/HOWTO/LVS-HOWTO_1.0-5.html#ss5.2
(it may not be terribly clear) then come up and ask again
Joe
--
Joseph Mack PhD, Senior Systems Engineer, Lockheed Martin
contractor to the National Environmental Supercomputer Center,
mailto:mack.joseph@epa.gov ph# 919-541-0007, RTP, NC, USA