Mailing List Archive

>From: rst@ai.mit.edu (Robert S. Thau)
>Date: Tue, 25 Jul 95 14:17:32 EDT
>
>Hmmm... this would allow for dieback beyond MaxServers when MaxServers
>has been reached, and seems the most sensible way to do it; the only
>problem would be deciding whether to worry about unlikely coincidences
>where a whole bunch of servers would die off all at once...

The solution to these worst-case scenarios is to introduce some randomness
into the server lifetimes.

David.

>From: rst@ai.mit.edu (Robert S. Thau)
>Date: Wed, 26 Jul 95 16:03:57 EDT
>As to votes, you want a vote you got it. Apache 0.8.2 is presently a
>drop-in replacement for NCSA 1.4.

I think this is the crux of the matter. I do not believe Apache 0.8.2
is a drop-in relacement for (i.e. compatible with) NCSA 1.4, as it is
different on a functional level; i.e. it has a different algorithm for
managing the sub-processes. Thus its config files must be semantically
different. [RobH may be of the same opinion.]

However, it is syntatically compatible with NCSA 1.4 config files, which
is presumably to what you were referring.

I think the only solution is to provide the Apache control parameters
with different names to the NCSA ones, and map the NCSA ones onto those.
e.g. Apache (maybe) has MinFreeServers and MaxFreeServers

If unset, then they default to some function of the NCSA
StartServers/MaxServers. However, I cannot imagine what function to use.

David.

Date: Thu, 27 Jul 95 14:36 BST
From: drtr@ast.cam.ac.uk (David Robinson)

I think this is the crux of the matter. I do not believe Apache 0.8.2
is a drop-in relacement for (i.e. compatible with) NCSA 1.4, as it is
different on a functional level; i.e. it has a different algorithm for
managing the sub-processes. Thus its config files must be semantically
different. [RobH may be of the same opinion.]

However, it is syntatically compatible with NCSA 1.4 config files, which
is presumably to what you were referring.

This distinction seems somewhat sensible.

I think the only solution is to provide the Apache control parameters
with different names to the NCSA ones, and map the NCSA ones onto those.
e.g. Apache (maybe) has MinFreeServers and MaxFreeServers

Well, as you've seen by now, I think 'Spare' might be a better way to
describe what these servers are ('Free' is perhaps more subject to
weird misinterpretations) and I'm thinking of making MaxServers just a
dummy instead of trying to set any of the Apache parameters. I do
hope this satisfies everyone and we can put this wrangle to bed.

I'm a bit calmer now, by the way --- being able to spend a whole day
on Real Work, and actually getting something concrete done on it,
helps quite a bit, though arguments like "your changes would break the
freeze but mine would not" are trying at the best of times. Apologies
to anyone to who I was short towards without good reason.

rst

While I don't feel it is appropriate for me to vote on
Apache implementation issues, I'd like to encourage you
to make very clear in some fashion that the Apache MinServers
and MaxServers are COMPLETELY different that the NCSA ones.
I don't think a name change is an unreasonable way to highlight
the distinction. No matter what you decide, I recommend the
creation of a document like

http://hoohoo.ncsa.uiuc.edu/docs/setup/Update.html

in which you highlight either the name change or that these
parameters need to be reset to values appropriate for the
Apache algorithm.

We've had a number of questions about these directives
that indicate a lot of people don't understand how they
work. As Apache becomes more popular, I suspect the
confusion will grow if the directive names are the same
but the effect on server performance is very different.

Elizabeth (Beth) Frank
NCSA Server Development Team
efrank@ncsa.uiuc.edu

From: efrank@ncsa.uiuc.edu (Elizabeth Frank)
Date: Fri, 28 Jul 1995 08:22:40 -0500
Precedence: bulk
Reply-To: new-httpd@hyperreal.com

While I don't feel it is appropriate for me to vote on
Apache implementation issues, I'd like to encourage you
to make very clear in some fashion that the Apache MinServers
and MaxServers are COMPLETELY different that the NCSA ones.
I don't think a name change is an unreasonable way to highlight
the distinction.

That seems to be the consensus (if it confuses people on this list,
the rest of the world doesn't stand a chance). Are MinSpareServers
and MaxSpareServers OK with everybody?

(The trick is to avoid names which are as misleading as "StartServers"
and "MaxServers", like, say, referring to what I'm calling
"MaxSpareServers" as "ServerPoolSize", even though the actual size of
the server pool can be considerably larger if the parameter is set low
and the server is under heavy load).

rst