Ocsar,
Does your Operton box have 2 gigabit NICS? Can linux and the box
drive 2 of them close to the max? Your utilitization implies that
it can. Very interesting. Can the PCI bus and backplane on this
box handle more traffic.
I just went through this with SUN. Solaris 8 is bottlnecked at 1Gps
period due the streams queue. Upgrading to Solaris 9 and using a V880,
I can get get about 700-800Mbps out of each NIC while they run together.
Finally with Solaris 9, we get past the 1Gbps barrier.
Of course the CPU utilization when driving the 1 or 2 NICs at full speed
is quite high. I can see how the V440 reaches a limit here.
Please post more Linux numbers for us.
Robert Maiello
Pioneer Data Systems
On Wed, 7 Dec 2005 02:41:40 +0100, Oscar Olsson <spam1 AT QBRANCH DOT SE> wrote:
>OK, after our first week with the new storage node, I thought I'd share
>some simple estimations of price/performance when it comes to the AMD
>Opteron, running Linux vs. the Sparc platform, running solaris.
>
>As you may be aware of, I have posted quite a few posts about NIC and
>server performance regarding the Sun V440 server, and the Cassini NIC. My
>conclusion was then that a V440 can push about 1gb/s of data from the
>network to the drives, before all four CPUs get maxed out.
>
>Our new 2CPU AMD opteron server can push 1gb data over the network with
>approx 50-60% CPU idle time, including I/O wait, system CPU, interrupts
>etc. I'm talking about actual production backup traffic flowing through
>networker here, not some silly benchmark test.
>
>The cost of a 4CPU V440 server is approx $25000. The cost of a HP DL385
>server with two opteron CPUs is about $6500. So, since the opteron can
>push twice as much data, the math looks like this: (25000*2)/6500=~7,69 .
>
>So, my conclusion is that you get about 8x performance for every invested
>dollar/euro/krona etc if you buy AMD based servers instead of sparc based
>ones. That's a significant price/performance ratio difference, worthy of
>most upper management people's attention.
>
>Ofcpurse, the Linux kernel/drivers could also have been a contributing
>factor, and that the OSes are not directly comparable. But I still see an
>obvious fact.. No wonder Sun is trying to market their AMD servers so
>hard! :)
>
>And why am I writing this? Partly because I wanted to follow up on my
>previous postnings, and partly because I think its in most sysadminst
>inetests to know what actual non-biased relevant performance figures may
>look like in the real world for the application they're probably running.
>
>//Oscar
>
>To sign off this list, send email to listserv AT listserv.temple DOT edu and
type "signoff networker" in the
>body of the email. Please write to networker-request AT listserv.temple DOT
>edu
if you have any problems
>wit this list. You can access the archives at
http://listserv.temple.edu/archives/networker.html or
>via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
>=========================================================================
To sign off this list, send email to listserv AT listserv.temple DOT edu and
type "signoff networker" in the
body of the email. Please write to networker-request AT listserv.temple DOT edu
if you have any problems
wit this list. You can access the archives at
http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
|
