httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chuck Murcko <>
Subject Re: project plan
Date Fri, 12 Jul 1996 22:12:19 GMT
Alexei Kosut liltingly intones:
> On Fri, 12 Jul 1996, Randy Terbush wrote:
> > Correct me if I'm wrong, but threading alone is not the answer to
> > performance issues.
> Nothing alone is the answer, but threading should make the server perform
> faster. It does away with forking entirely, for one thing (remember the
> performance boost from 0.6 to 0.8, or NCSA 1.3 to 1.4?), which not only
> speeds execution, it dramatically reduces CPU and memory usage, which
> tends to have an uplifting performance benefit.
> Spyglass Server still claims they're something like seven times faster
> than Apache. All that speed has to come from somewhere, and I don't think
> they bundle an extra CPU chip with each of their servers (though for
> $15,000, they might as well).
Weren't Spyglass the folks with the keepalive-rigged benchmark? Or am I
mistaking them for someone else?

It's also important to remember that the real world latency time for
clients averages somewhere around 200-250 ms, because many people are
on dialups. Things like kernel patches to prevent errant slow start,
increasing mbuf size to 256 bytes where necessary to prevent multi-
segment client requests, handling connections stuck in LAST_ACK state
because of broken clients, and switching over to T/TCP eventually will
help there, as well as on 10 Mbit/sec networks.

I'm not denigrating John's observations, just pointing out that the
whole world isn't yet at Ethernet speed. His points are also valid,
given that he doesn't have Netscape Navigator to work around. 8^)

Chuck Murcko	N2K Inc.	Wayne PA
And now, on a lighter note:
Westheimer's Discovery:
	A couple of months in the laboratory can frequently save a
couple of hours in the library.

View raw message