Jason Spisak wrote:

You bet, Christopher.

Linux RH 5.2 Python 1.5.2 Netscape 4.61 & 4.08 (Both were unable to get a response from ZServer) ZServer with no Apache No custom configurations, just the default installed from beta5 download Using the same Data.fs file from beta2, beta3, beta4, now beta5 Process took the keyboard interupt Restarted and everything is fine(For now).

Behavior is idential from the outside to the behavior in beta4 where the server would just stop giving back responses/taking requests. The browser just says "host contacted waiting for reply..." and the icon keeps churning.

Did you try making other browser requests while this was happening? For example, did ou click the stop button and try to visit another page or to visit another page from another browser window? Do you remember anything about what you were doing before this happened? How long had Zope been running? What sorts of objects were you using? Was the Zope process consuming significant CPU while it was not responding? Jim -- Jim Fulton mailto:jim@digicool.com Python Powered! Technical Director (888) 344-4332 http://www.python.org Digital Creations http://www.digicool.com http://www.zope.org Under US Code Title 47, Sec.227(b)(1)(C), Sec.227(a)(2)(B) This email address may not be added to any commercial mail list with out my permission. Violation of my privacy with advertising or SPAM will result in a suit for a MINIMUM of $500 damages/incident, $1500 for repeats.

On Wed, Aug 25, 1999 at 12:09:31PM -0400, Damian Morton wrote:

What I havent seen discussed in this list at all is performance. Is ZOPE ready for industrial strength usage? Can it truly compete with Vignette, Broadvision, etc etc (apart from on price?). Will ZOPE benefit from SMP? Has anyone run ZOPE clusters?

1. Don't know about Vignette, Broadvision etc. 2. Zope should benefit from SMP as it's multithreaded (even with that global interpreter lock it still should do fine) 3. Ask the digicool people to tell you more about ZEO if you want to build a Zope cluster. 4. A small benchmark to see how Zope scales under high load (yes I know benchmarks are irrelevant, but...): Server: AMD K6-2/333MHz w/128M RAM running Linux (upgraded RedHat 6.0, kernel 2.2.12-final3), Python 1.5.2 and a couple of other services which were idle in the moment of the benchmark. UDMA IDE disk which can get ~10M/sec. Latest CVS Zope running with ZServer (no apache or pcgi in the middle). Client: same as server but with 64M RAM (Linux, same distribution, same kernel) running ApacheBench instructed to do 1000 requests (100 at once). Network: 100Mbps, almost idle during the benchmark. Results: when requesting the default installation welcome page it gets ~46 hits/second. With the more complicated QuickStart page it does ~10 hits/second. The *really* interesting part was when I made another test with conecurrency level=1000. The results were the same, so ZServer apparently does not have scalability problems. -Petru

On Wed, 25 Aug 1999, Damian Morton wrote:

Hmm, 10 hits per second doesnt really seem that fast to me - especially Well. It's 10 hits for a really complicated page. (The tree tag has to activate all subobjects and collect the layout from them.)

So you should compare it more to .ASP, .cfm, .cgi or whatever mechanism page that does a query (in the case of the QuickStart document in Zope this is an object database, but it could also come from an SQL server.)

given that here were no real bandwidth limitations in the test. That works out to about a million hits a day. Well, some observations: 1.) 1-4 million hits per day are a nice site. How many sites do have this traffic? 2.) How many of these sites are implemented on an aging K6-233 with one IDE drive? (not even on a really fast IDE drive) More interesting for such a site would be a benchmark on K7-650 with more RAM and an U2W RAID array. 3.) The benchmark was out of the box: No tuning whatsoever. For example, ZServer by default starts ``only'' 4 threads. So with 100 concurrent requests 96 requests are queued only. One would have to experiment to tune it a bit better. 4.) Another tuning sometimes possible would be a server-side squid cache.

With the concurrency level thing. Are you saying that in the first test, you made 1000 requests, with 100 active at any given time, but then you made a second test in which you made 1000 requests all at once, and the results were that the server managed 10 hits per second on the more complicated page, i.e. it took about 100 seconds for the server to process both tests. That's what I understood.

What is ZEO? Z Enterprise Objects? I find no reference to ZEO anywhere on If I may guess, this will refer to DC technique of linking ZOPE servers. (This is practical not only for clustering stuff: Consider having your database on MS-SQL NT server. Now, your Webserver is a Linux box. How do you access the database? Well, by having another smaller ZOPE server on the NT box, which accesses MS-SQL via ODBC, and is accessed from the external ZOPE server.)

Side note: I've found, that the digicool website offers almost no navigatable content about the commercial products DC has to offer :( Andreas -- Andreas Kostyrka | andreas@mtg.co.at phone: +43/1/7070750 | phone: +43/676/4091256 MTG Handelsges.m.b.H. | fax: +43/1/7065299 Raiffeisenstr. 16/9 | 2320 Zwoelfaxing AUSTRIA

On Wed, 25 Aug 1999, Jason Spisak wrote:

Is there any other things pseides the process data that I can check while this is happening next time? If this is on Linux, you could try to do strace -p <pid_of_zope_process> 2>/tmp/crazy_zope.log

Andreas -- Andreas Kostyrka | andreas@mtg.co.at phone: +43/1/7070750 | phone: +43/676/4091256 MTG Handelsges.m.b.H. | fax: +43/1/7065299 Raiffeisenstr. 16/9 | 2320 Zwoelfaxing AUSTRIA