How small a box can zope run on?
Hi, Just a quick question: We are building some device controller software and to control those we have a normal 486 with 24MB RAM running linux. Has anybody tried running zope on such a modest setup? There is only going to be one or two users connected to Zope at once, so we don't need much performance in terms of page views per second :) Cheers, Alexander.
I have had it running on a Pentium 133 w/ 16 MB of RAM and it was actually doing ok. Not quick, per se, but ok. J
From: "Alexander Limi" <alexander@limi.net> Date: Wed, 25 Oct 2000 18:43:57 +0200 To: <zope@zope.org> Subject: [Zope] How small a box can zope run on?
Hi,
Just a quick question:
We are building some device controller software and to control those we have a normal 486 with 24MB RAM running linux. Has anybody tried running zope on such a modest setup? There is only going to be one or two users connected to Zope at once, so we don't need much performance in terms of page views per second :)
Cheers,
Alexander.
_______________________________________________ Zope maillist - Zope@zope.org http://lists.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://lists.zope.org/mailman/listinfo/zope-announce http://lists.zope.org/mailman/listinfo/zope-dev )
On Wed, 25 Oct 2000 18:43:57 +0200, "Alexander Limi" <alexander@limi.net> wrote:
We are building some device controller software and to control those we have a normal 486 with 24MB RAM running linux. Has anybody tried running zope on such a modest setup? There is only going to be one or two users connected to Zope at once, so we don't need much performance in terms of page views per second :)
Yes, Im using Zope in a similar scenario and it works well. I found a benefit in using a smaller number of publisher threads (I use -T 2 ). Each thread gets its own copy of the ODB, which is an unnecessary memory hog for the expected load. Toby Dickenson tdickenson@geminidataloggers.com
At 10:15 am +0100 26/10/00, Toby Dickenson wrote:
On Wed, 25 Oct 2000 18:43:57 +0200, "Alexander Limi" <alexander@limi.net> wrote:
We are building some device controller software and to control those we have a normal 486 with 24MB RAM running linux. Has anybody tried running zope on such a modest setup? There is only going to be one or two users connected to Zope at once, so we don't need much performance in terms of page views per second :)
Yes, Im using Zope in a similar scenario and it works well.
I found a benefit in using a smaller number of publisher threads (I use -T 2 ). Each thread gets its own copy of the ODB, which is an unnecessary memory hog for the expected load.
????!!! This is the first I've ever seen written on this subject. If it's correct (and I have no reason to disbelieve Toby - I've tried a few things on our box here and it seems to pan out), then this explains large memory consumption on our servers (roughly 80 - 180 Mbyte RAM used per server). I shall be doing a sweep of our Zope servers and packing the databases (the ODBs that is). One thing though, is there a 'party line' on what number the -T parameter should be set to? Coming from an Apache background, I'm used to having dozens of httpd processes hanging around waiting for something to happen. I've always assumed that threads did much the same thing. Perhaps I'm wrong.... thoughts? Tone ------ Dr Tony McDonald, FMCC, Networked Learning Environments Project http://nle.ncl.ac.uk/ The Medical School, Newcastle University Tel: +44 191 222 5116 A Zope list for UK HE/FE http://www.fmcc.org.uk/mailman/listinfo/zope
Tony McDonald wrote:
This is the first I've ever seen written on this subject. If it's correct (and I have no reason to disbelieve Toby - I've tried a few
There was a good discussion about this a couple of months ago. Check the archives for: zope2.2.0 and what is high load -- Tim Cook -- Cook Information Systems | Office: (901) 884-4126 8am-5pm CDT Free Practice Management | http://FreePM.org Censorship: The reaction of the ignorant to freedom.
Toby Dickenson wrote:
On Wed, 25 Oct 2000 18:43:57 +0200, "Alexander Limi" <alexander@limi.net> wrote:
We are building some device controller software and to control those we have a normal 486 with 24MB RAM running linux. Has anybody tried running zope on such a modest setup? There is only going to be one or two users connected to Zope at once, so we don't need much performance in terms of page views per second :)
Yes, Im using Zope in a similar scenario and it works well.
I found a benefit in using a smaller number of publisher threads (I use -T 2 ). Each thread gets its own copy of the ODB, which is an unnecessary memory hog for the expected load.
What OS is this on? On Linux each thread does _NOT_ get a copy of the ODB. It just _looks_ like it. See the archives for details. The benefit from smaller thread counts is that: A) Multiple threads is not a big boost on uniprocessor machines B) Python does some different things with threads, see the Global Interpreter Lock for details. Changin the realted value in Zope will acheive differing levels of performance. -- E PLURIBUS LINUX
On Fri, 27 Oct 2000 09:39:54 -0600, Bill Anderson <bill@noreboots.com> wrote:
What OS is this on?
All of them ;-)
On Linux each thread does _NOT_ get a copy of the ODB. It just _looks_ like it.
I suspect you are referring to the characteristic that several Linux memory-reporting tools list the memory used by one Zope process once for each thread. That's not the characteristic I am referring to....Each zope publisher thread really does have its own copy of the ZODB object cache. You can verify this by checking the value "Total number of objects in all of the caches combined" from Control Panel. This number should be roughly (number of threads) * (target size), although there are many factors that can affect it.
See the archives for details. The benefit from smaller thread counts is that: A) Multiple threads is not a big boost on uniprocessor machines
This is only true if Zope is saturating your processor. It may not be true if you are publishing any methods that are mostly I/O (file access, or other web requests)
B) Python does some different things with threads, see the Global Interpreter Lock for details. Changin the realted value in Zope will acheive differing levels of performance.
Yes Toby Dickenson tdickenson@geminidataloggers.com
Toby Dickenson wrote:
On Fri, 27 Oct 2000 09:39:54 -0600, Bill Anderson <bill@noreboots.com> wrote:
What OS is this on?
All of them ;-)
On Linux each thread does _NOT_ get a copy of the ODB. It just _looks_ like it.
I suspect you are referring to the characteristic that several Linux memory-reporting tools list the memory used by one Zope process once for each thread.
That's not the characteristic I am referring to....Each zope publisher thread really does have its own copy of the ZODB object cache.
of the Object CACHE, that I can by, of the ZODB itsself, no.
You can verify this by checking the value "Total number of objects in all of the caches combined" from Control Panel. This number should be roughly (number of threads) * (target size), although there are many factors that can affect it.
See the archives for details. The benefit from smaller thread counts is that: A) Multiple threads is not a big boost on uniprocessor machines
This is only true if Zope is saturating your processor. It may not be true if you are publishing any methods that are mostly I/O (file access, or other web requests)
I am only talking about Zope's use of it here. _Zope_ doesn't gain much from multiple CPUs. ( I have machines here that have _many_ processors, and only two that have less than two ) -- E PLURIBUS LINUX
participants (6)
-
Alexander Limi -
Bill Anderson -
J. Atwood -
Tim Cook -
Toby Dickenson -
Tony McDonald