Mailing List Archive

On Mar 15, 12:06pm, David Robinson wrote:
} Subject: Re: CERN
} >A few people that I really respect at work (sun) are suggesting (quite
} >strongly in fact) that we look at using CERN as a base.
} I thought the CERN server was really huge, with lots of features we
} don't want, like proxying. I don't actually know of anyone using the
} CERN server for external accesses, only as a proxy gateway for
} internal users.

Sun's www.sun.com is CERN, as are all the proxy's.

}
} >I am also talking to some marketers at Sun about some hardware. I
} >think we can really use a nice MP system to test the server. :-)
} Sounds very nice...
} Pro tem we have several MP suns here which I can test code on.
} The only problem is finding an unloaded one which is not trying to
} simulate a galaxy star by star...

Great...how much load can we generate against them when in test mode?

> A few people that I really respect at work (sun) are suggesting (quite
> strongly in fact) that we look at using CERN as a base. I know that
> the CERN server has been hacked to include all the NSCA features by
> folks here and I was told it was not difficult. What are peoples
> thoughts on this? [ as cliff ducks ]

The CERN server is larger than NCSA (at least, larger than NCSA without
the HUGE_ patch :-) and considerably slower.

> I am also talking to some marketers at Sun about some hardware. I
> think we can really use a nice MP system to test the server. :-)

I have a ten processor SC2000 system with 1.5GB of RAM.

It doesn't have a web server running on it right now but I could
certainly install one.

Our web server (www.msstate.edu) is a dual processor SS20 with 256MB
of memory (temporarily). It has the advantage of being an active
server (US mirror for the internet movie database). So when we have
a stable enough server to put into a production environment that
will be an active test case.

Between the two I think we have Sun MP well covered.

Date: Wed, 15 Mar 1995 03:56:12 -0800
From: Cliff Skolnick <cliffs@steam.com>
Precedence: bulk
Reply-To: new-httpd@hyperreal.com


A few people that I really respect at work (sun) are suggesting (quite
strongly in fact) that we look at using CERN as a base. I know that
the CERN server has been hacked to include all the NSCA features by
folks here and I was told it was not difficult. What are peoples
thoughts on this? [ as cliff ducks ]

Cliff

The CERN server code is enormous, and I find it difficult to read.
Once the obvious performance problems with the NCSA server are fixed
(P1, P2, P3), it also tests as being substantially slower.

Also, large portions of the server --- most of the content arbitration
code, for instance --- are actually part of the common code library;
any changes to this code (to add negotiation based on "level=", for
instance) would have to be a change to the *library*.

This means we'd have to either decide to maintain our own version of
the library as well, or coordinate very carefully with W3O --- since
that the W3O's own current development versions of the library don't
work with the server at all, this could be extremely tricky.

Finally, when you put the server code and library code together, the
combined code base we'd be managing has over 7 times as many lines as
the NCSA code base. It's pretty unwieldly. I'd just as soon do
without it.

rst

>A few people that I really respect at work (sun) are suggesting (quite
>strongly in fact) that we look at using CERN as a base.
I thought the CERN server was really huge, with lots of features we
don't want, like proxying. I don't actually know of anyone using the
CERN server for external accesses, only as a proxy gateway for
internal users.

>I am also talking to some marketers at Sun about some hardware. I
>think we can really use a nice MP system to test the server. :-)
Sounds very nice...
Pro tem we have several MP suns here which I can test code on.
The only problem is finding an unloaded one which is not trying to
simulate a galaxy star by star...

> From: Cliff Skolnick <cliffs@steam.com>

> A few people that I really respect at work (sun) are suggesting (quite
> strongly in fact) that we look at using CERN as a base. I know that
> the CERN server has been hacked to include all the NSCA features by
> folks here and I was told it was not difficult. What are peoples
> thoughts on this? [ as cliff ducks ]
>
> Cliff

rst:
> The CERN server code is enormous, and I find it difficult to read.
> Once the obvious performance problems with the NCSA server are fixed
> (P1, P2, P3), it also tests as being substantially slower.
>

[snip]

> Finally, when you put the server code and library code together, the
> combined code base we'd be managing has over 7 times as many lines as
> the NCSA code base. It's pretty unwieldly. I'd just as soon do
> without it.
>
> rst
>

We can barely manage an NCSA based project, there's no chance of us being able to
run a CERN one. It'd be great to have the choice but we don't. Stick to NCSA
and keep a note of any general lessons learned. If we're up to it, then things
we do for NCSA should be implementable in a CERN code-stylee (er?). The MAIN
reason we chose NCSA as a base was because it was relatively small and neat code
and we all used it every day. The same can't be said of CERN, even though it may
be a better system on-paper.

Please don't switch to CERN till we've exhausted the NCSA route.

Ay.