This is moved to the Zope list from private email. Brian Takashi Hooper wrote:
Hi Chris, thanks for the quick answer!
On Thu, 03 Feb 2000 21:54:19 -0500 Chris McDonough <chrism@digicool.com> wrote: (BTW, go home, it's late :-) )
[snip]
So as far as overrunning your available Sybase connections, I don't think you have to sweat it much as long as you have at least 7 available licensed connections and you've only got one connection to the database within Zope. I'm not really quite sure why you're only seeing four threads active in a -t 20 environment, however. It may have something to do with the timing of the requests as they come in to Zope via ZServer (Zope may only need four threads to serve back the data even though you say your client runs with six). There's a lot of stuff going on in there. :-) I'll say - well I certainly didn't know anything about the magic 7 number limit (after I knew what to search for, I did find the magic limit in ZODB/DB.py), that's very interesting.
I am still wondering about my application threads, tho. I have been loading my server using a threaded client with gradually increasing numbers of threads... basically what I've been doing is increasing the number of virtual clients until my request latency goes significantly up, and vmstat on the server end indicates things are pretty busy. Then I started looking at thread.get_ident() in ZServer/PubCore/ZServerPublisher, something like:
import creosote, thread
class ZServerPublisher: def __init__(self, accept): while 1: try: name, request, response=accept() creosote.spew('REQUEST handled by: %d' % thread.get_ident()) ^^^^^^^^^^^^^ publish_module( name, request=request, response=response) finally: response._finish() request=response=None
I don't know if this way of looking at things makes sense or not, obviously Heisenberg's principle applies...
Anyway I (of course) get something like this if I listen to UDP on 7739:
REQUEST handled by: 3076 REQUEST handled by: 2051 REQUEST handled by: 3076 REQUEST handled by: 2051 REQUEST handled by: 2051 REQUEST handled by: 2051 REQUEST handled by: 2051 REQUEST handled by: 2051 REQUEST handled by: 3076 REQUEST handled by: 2051 REQUEST handled by: 2051 REQUEST handled by: 2051 REQUEST handled by: 3076 REQUEST handled by: 2051 REQUEST handled by: 3076 REQUEST handled by: 2051 REQUEST handled by: 2051
Looking at things via vmstat, small requests seem to be CPU bound; large requests seem to be I/O bound; I guess this makes sense...
However, no matter what I try (requesting large (>50K) documents, small documents (<1K), etc.) I can't seem to visualize more than 3 threads at once. Am I looking at this in totally the wrong way? Assuming 7 is the magic pool limit, then it seems like I would want to try to get things up to 7 simultaneous RDB connections if I'm serving out all RDB-using pages.
Any ideas?
--Brian
P.S. Should we move this discussion to the Zope list? Maybe someone else out there is wondering the same thing, or better yet already knows the answer?