how many do I need, and how will I know? -- -- Einar Næss Jensen http://einar.nidelven-it.no/einarblog http://www.homemade.no tlf: +47 90990249
I just spent a few minutes googling it and failed, but I thought I read on the Zope wiki that for higher trafic sites, it was better to have bigger caches (50,000+) and fewer threads (2). Jake _______________________ http://www.ZopeZone.com On Mon, January 30, 2006 5:47 pm, Jens Vagelpohl said:
On 30 Jan 2006, at 22:43, Einar Næss Jensen wrote:
how many do I need, and how will I know?
You do not need to worry about it unless you have a really unusual setup. The default is fine for 99.9% of all situations.
jens
_______________________________________________ Zope maillist - Zope@zope.org http://mail.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://mail.zope.org/mailman/listinfo/zope-announce http://mail.zope.org/mailman/listinfo/zope-dev )
http://plone.org/documentation/how-to/threads-vs-cache-size There it was. Jake _______________________ http://www.ZopeZone.com On Tue, January 31, 2006 9:59 am, Jake said:
I just spent a few minutes googling it and failed, but I thought I read on the Zope wiki that for higher trafic sites, it was better to have bigger caches (50,000+) and fewer threads (2).
Jake _______________________ http://www.ZopeZone.com
On Mon, January 30, 2006 5:47 pm, Jens Vagelpohl said:
On 30 Jan 2006, at 22:43, Einar Næss Jensen wrote:
how many do I need, and how will I know?
You do not need to worry about it unless you have a really unusual setup. The default is fine for 99.9% of all situations.
jens
_______________________________________________ Zope maillist - Zope@zope.org http://mail.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://mail.zope.org/mailman/listinfo/zope-announce http://mail.zope.org/mailman/listinfo/zope-dev )
_______________________________________________ Zope maillist - Zope@zope.org http://mail.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://mail.zope.org/mailman/listinfo/zope-announce http://mail.zope.org/mailman/listinfo/zope-dev )
On 31 Jan 2006, at 15:01, Jake wrote:
That article contains little information to back up the conclusions, and some of it is patently wrong. Another case of hearsay and half- thruths being propagated by well-meaning but uninformed parties. jens
Jens Vagelpohl wrote at 2006-1-31 15:07 +0000:
On 31 Jan 2006, at 15:01, Jake wrote:
That article contains little information to back up the conclusions, and some of it is patently wrong. Another case of hearsay and half- thruths being propagated by well-meaning but uninformed parties.
There is a convincing example (I think from Matt Kromer) what happens to "mean response time" when the number of threads is increased from 1 to 2: Assume you have 2 units of work. If you have a single thread, 1 unit is done, then the second. The mean response time is (1 + 2) / 2 = 1.5. Now assume you have two threads (but a single processor). Both threads work (quasi) parallel but have only half of the processor power. Both will take 2 time units to perform their unit of work. The mean response time is (2 + 2) / 2 = 2. You get the same ratios whenever you double the number of threads (but keep the number of processors and all your work is CPU bound). Things become more difficult when you have a mix of very small requests and very long requests. You then may want a higher level of concurrency to not let the small requests being blocked by the long ones. -- Dieter
Am I gleaming from that that you are proposing a less-is-more approach to threads? Here is what I have been using: - Zope 2.7.8 (Plone 2.1.2) - RH Linux - AMD Athlon 64 3200+ 2.0 GHz - 2GB DDR RAM - 120,000 hits a day - 392,036 objects in database - 2 threads - 100,000 object cache size This seems to eat up about half of the RAM on the server. Would one thread but double the cache size do better? Jake _______________________ http://www.ZopeZone.com On Tue, January 31, 2006 2:40 pm, Dieter Maurer said:
Jens Vagelpohl wrote at 2006-1-31 15:07 +0000:
On 31 Jan 2006, at 15:01, Jake wrote:
That article contains little information to back up the conclusions, and some of it is patently wrong. Another case of hearsay and half- thruths being propagated by well-meaning but uninformed parties.
There is a convincing example (I think from Matt Kromer) what happens to "mean response time" when the number of threads is increased from 1 to 2:
Assume you have 2 units of work. If you have a single thread, 1 unit is done, then the second. The mean response time is (1 + 2) / 2 = 1.5.
Now assume you have two threads (but a single processor). Both threads work (quasi) parallel but have only half of the processor power. Both will take 2 time units to perform their unit of work. The mean response time is (2 + 2) / 2 = 2.
You get the same ratios whenever you double the number of threads (but keep the number of processors and all your work is CPU bound).
Things become more difficult when you have a mix of very small requests and very long requests. You then may want a higher level of concurrency to not let the small requests being blocked by the long ones.
-- Dieter _______________________________________________ Zope maillist - Zope@zope.org http://mail.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://mail.zope.org/mailman/listinfo/zope-announce http://mail.zope.org/mailman/listinfo/zope-dev )
On 1/31/06, Jake <jake@zopezone.com> wrote:
Am I gleaming from that that you are proposing a less-is-more approach to threads?
Here is what I have been using: - Zope 2.7.8 (Plone 2.1.2) - RH Linux - AMD Athlon 64 3200+ 2.0 GHz - 2GB DDR RAM - 120,000 hits a day - 392,036 objects in database - 2 threads - 100,000 object cache size
This seems to eat up about half of the RAM on the server.
Would one thread but double the cache size do better?
Jake _______________________ http://www.ZopeZone.com
On Tue, January 31, 2006 2:40 pm, Dieter Maurer said:
Jens Vagelpohl wrote at 2006-1-31 15:07 +0000:
On 31 Jan 2006, at 15:01, Jake wrote:
That article contains little information to back up the conclusions, and some of it is patently wrong. Another case of hearsay and half- thruths being propagated by well-meaning but uninformed parties.
There is a convincing example (I think from Matt Kromer) what happens to "mean response time" when the number of threads is increased from 1 to 2:
Assume you have 2 units of work. If you have a single thread, 1 unit is done, then the second. The mean response time is (1 + 2) / 2 = 1.5.
Now assume you have two threads (but a single processor). Both threads work (quasi) parallel but have only half of the processor power. Both will take 2 time units to perform their unit of work. The mean response time is (2 + 2) / 2 = 2.
You get the same ratios whenever you double the number of threads (but keep the number of processors and all your work is CPU bound).
Things become more difficult when you have a mix of very small requests and very long requests. You then may want a higher level of concurrency to not let the small requests being blocked by the long ones.
-- Dieter _______________________________________________ Zope maillist - Zope@zope.org http://mail.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://mail.zope.org/mailman/listinfo/zope-announce http://mail.zope.org/mailman/listinfo/zope-dev )
_______________________________________________ Zope maillist - Zope@zope.org http://mail.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://mail.zope.org/mailman/listinfo/zope-announce http://mail.zope.org/mailman/listinfo/zope-dev )
I would recommend exploring ZEO and Squid if you're *this* worried about performance. There's only so much that one machine can do. If I understand the Zope framework, the idea is that read-to-write ratio for the ZODB is going to be much greater than one, meaning that there are going to be a lot more reads than writes. In this case, having some load balancing and ZEO client caching will make overall performance faster. However, if you're doing lots of writing to the ZODB, you'll probably need to explore ZEO even more in the area of mounting different ZODBs at different places in your ZEO clients' trees. Like some of the other people said, you probably should just tune it as needed. It's also a Bad Thing to be running your servers at or near capacity; you should leave plenty of headspace for usage spikes. I've certainly run into situations where Zope stuff was processor-bound and slowing everything down, but the RAM wasn't even 25% full. In cases like this, more zserver threads (or more ZEO clients) helps even things out. YMMV. -- Floyd May Senior Systems Analyst CTLN - CareerTech Learning Network fmay@okcareertech.org
Jake wrote at 2006-1-31 15:53 -0500:
Am I gleaming from that that you are proposing a less-is-more approach to threads?
That's something proposed several times in the list. In our installation, we used the default (4 threads) for a long time -- until we met overload situations (all 4 threads in use over long periods of time) when we switched to 6 threads. However, in our situation not only Zope handles the application. Beside that there are backend resources such as various databases that can work concurrently with Zope. Therefore, our application is not CPU bound (at least not on the host, Zope runs on) and increasing the number of threads can increase throughput.
... This seems to eat up about half of the RAM on the server.
Would one thread but double the cache size do better?
You have read my message to the end (and Florent's comment)? If you have a single thread, then a single long running request would exclude all other (maybe very short) requests until it finisches. I am convinced that you want some degree of concurrency to favour short running requests. 4 threads is not too a default. -- Dieter
Dieter Maurer wrote:
Jens Vagelpohl wrote at 2006-1-31 15:07 +0000:
On 31 Jan 2006, at 15:01, Jake wrote:
http://plone.org/documentation/how-to/threads-vs-cache-size That article contains little information to back up the conclusions, and some of it is patently wrong. Another case of hearsay and half- thruths being propagated by well-meaning but uninformed parties.
There is a convincing example (I think from Matt Kromer) what happens to "mean response time" when the number of threads is increased from 1 to 2:
Assume you have 2 units of work. If you have a single thread, 1 unit is done, then the second. The mean response time is (1 + 2) / 2 = 1.5.
Now assume you have two threads (but a single processor). Both threads work (quasi) parallel but have only half of the processor power. Both will take 2 time units to perform their unit of work. The mean response time is (2 + 2) / 2 = 2.
This is specious reasoning from a statistical point of view. You never have only two requests. To have a proper statistical evaluation, you need a theoretical model of the distribution of requests over time. This can be as simple as "a request every t". Then you have a continuous flow of requests, and the two means become equal. Just try it with 4 units of work and equivalent time distribution as in the original example: - Case 1: (1+2+3+4)/4 = 2.5 - Case 2: (2+2+4+4)/4 = 3 If you generalize to n units of work, you'll get (the math is simple): - Case 1: n(n+1)/2n = n/2+0.5 - Case 2: 4((n/2)(n/2+1)/2)/n = n/2+1 Which are equivalent to n/2 when n grows. Florent -- Florent Guillaume, Nuxeo (Paris, France) Director of R&D +33 1 40 33 71 59 http://nuxeo.com fg@nuxeo.com
On 31 Jan 2006, at 14:59, Jake wrote:
I just spent a few minutes googling it and failed, but I thought I read on the Zope wiki that for higher trafic sites, it was better to have bigger caches (50,000+) and fewer threads (2).
The term "high traffic site" doesn't mean a thing when it comes to threads or the cache (or in general). It's way too general. Determinations about cache sizes or thread numbers can only be made with specific requirements, (estimated) traffic data, hardware specifications and knowledge about the Zope application in hand, period. Less threads and bigger caches only make sense when you are under memory pressure and (hopefully) your pages render fast enough so that less threads can still satisfy the rendering load. jens
It would be nice if someone post some guidelines for threads and cache size, like a table. RAM / Hits / Threads / Cache Size 2 GB / 20,000,000 / 3 / 50,000 5 GB / 20,000,000 / 5 / 100,000 Jake _______________________ http://www.ZopeZone.com On Tue, January 31, 2006 10:05 am, Jens Vagelpohl said:
On 31 Jan 2006, at 14:59, Jake wrote:
I just spent a few minutes googling it and failed, but I thought I read on the Zope wiki that for higher trafic sites, it was better to have bigger caches (50,000+) and fewer threads (2).
The term "high traffic site" doesn't mean a thing when it comes to threads or the cache (or in general). It's way too general. Determinations about cache sizes or thread numbers can only be made with specific requirements, (estimated) traffic data, hardware specifications and knowledge about the Zope application in hand, period.
Less threads and bigger caches only make sense when you are under memory pressure and (hopefully) your pages render fast enough so that less threads can still satisfy the rendering load.
jens
_______________________________________________ Zope maillist - Zope@zope.org http://mail.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://mail.zope.org/mailman/listinfo/zope-announce http://mail.zope.org/mailman/listinfo/zope-dev )
--On 31. Januar 2006 12:32:35 -0500 Jake <jake@zopezone.com> wrote:
It would be nice if someone post some guidelines for threads and cache size, like a table.
RAM / Hits / Threads / Cache Size 2 GB / 20,000,000 / 3 / 50,000 5 GB / 20,000,000 / 5 / 100,000
Such a table is pretty much worthless. A ZEO cache of 50.000 objects tells you nothing about the RAM to be used. The cache could hold 50.000 small objects or 50.000 very large objects eating all your RAM. The relation between #threads and hits also depends very much on your application. To make it short: dealing with larger installation always requires some iterations of oberserving your system and tuning it. -aj
Well, I have been using Zope for over 6 years and I still don't think I have mastered what is truly best other than get a good server, with a lot of RAM and bump up your threads * cache to use up about 50% of it. Jake _______________________ http://www.ZopeZone.com On Tue, January 31, 2006 12:46 pm, Andreas Jung said:
--On 31. Januar 2006 12:32:35 -0500 Jake <jake@zopezone.com> wrote:
It would be nice if someone post some guidelines for threads and cache size, like a table.
RAM / Hits / Threads / Cache Size 2 GB / 20,000,000 / 3 / 50,000 5 GB / 20,000,000 / 5 / 100,000
Such a table is pretty much worthless. A ZEO cache of 50.000 objects tells you nothing about the RAM to be used. The cache could hold 50.000 small objects or 50.000 very large objects eating all your RAM. The relation between #threads and hits also depends very much on your application. To make it short: dealing with larger installation always requires some iterations of oberserving your system and tuning it.
-aj
participants (7)
-
Andreas Jung -
Dieter Maurer -
Einar Næss Jensen -
Florent Guillaume -
Floyd May -
Jake -
Jens Vagelpohl