Dan Hollis wrote:
>
> While we're on the subject of HP performance, I'd like to draw
> attention to a little HTTP benchmark a couple of us have been
> doing for the past few months. You can see the results at:
>
> http://www.anime.net/httpd_benchmarks.html
>
> Keep in mind this benchmark is not intended to show
> "average usage", it is designed to benchmark the
> _upper bound_ of the web server. Any flames regarding
> the testing methodology will be cheerfully circular filed ;-)
I haven't checked Dan's "official" page above yet, but he may have my figures
on NCSA httpd 1.3 and the OpenMarket server on 5.0. Dan can now file this in
his circular file or /dev/null :-) but I'd like to put my $0.02 in...
Dan's figures are likely representative for Dan's observations. Mine were
from a different environment, different pages, etc. These benchmarks are
like MIPS, they're OK relative to a common environment, but not relative to
any other environment (not necessarily).
He has some "smoking" figures for some servers, but they're run in loopback
(no network latency). You can get high hits/sec figures with small pages and
high Kb/sec with large pages, so these can be tailored as well. Also omitted
was the concurrency levels, the size of the server pool (where applicable),
and other *very* influential factors. Not that he was trying to bias the
figures (don't get that impression) just saying "your mileage may vary".
I'll sit here and tell you that a 386/33 PC running WinHTTPD can smoke a 950
in httpd performance particularly compared with NCSA httpd. The freeware
NCSA httpd port works, but unless you have a higher-end system, it can't even
compete with a lowly PC. At the other end of the spectrum, HP has officially
stated that the OpenMarket server is <at least> 5x faster than the NCSA port.
I will tell you that on a 950 it was well beyond 10x faster than NCSA and it
took all of 15 minutes to install and have it serve our existing pages.
I can't speak for a high-end 3000 (the 960 is the best I have at the moment)
but I suspect they could be competitive with OpenMarket compared to other
platforms. The 960 running OpenMarket does smoke the 386 WinHTTPD server,
but our Suns running NCSA 1.5 (not ported to MPE due to BSD socket quirk that
has not yet been fixed). As I've said before, MPE's Posix doesn't fork()
well at all, and that is what kills the freeware NCSA httpd port. OpenMarket
uses a server pool and posix threads and doesn't have to fork() for simple
connections although it can be burdened by cgi applications.
I suppose what I'm trying to say is that Netscape would not want to replace
their home.netscape.com machine with any 3000 (millions of hits). Otherwise,
the 3000 is a VERY viable web server running OpenMarket on any model of
hardware. But your performance with the NCSA port might leave something to
be desired on lower-end machines.
This isn't a 3000 and/or Posix weakness in a general sense. Web servers (and
clients for that matter) exploit every known weakness of every relevant
system-dependent implementation of networking/file systems/etc in a manner
previously unheard of. Netscape allows numerous parallel TCP connections,
most of which are of very brief duration; TCP was never designed for such a
load nor the error handling involved when you click the "stop" icon to abort
a transfer. It is the ultimate stress test, and not just HP3000 related.
Trumpet Winsock (earlier releases), WRQ's Reflection Network Series, and
numerous other vendors have spent their quota of cursing over http problems
(client and server).
I'm digressing :-) I would like to see some "apples to apples" comparison
tests of the web benchmarks and I am perhaps at fault for not providing the
3000 figures sooner, but I can't elaborate on why I haven't (trust me). I
will attack the issue again after HPWorld.
Dan - can you supply the zbench package and actual benchmark pages on your
web site so that those of us who can contribute numbers can obtain "equal
ground" environment for comparison?
Jeff Kell <[log in to unmask]>
|
