John Morton wrote:

Most of that memory is actually shared between the threads, so the size of one thread is closer to your real memory usage.

This is one area that at first was very confusing, the way Linux reports memory appears to be inconsistent when it comes to memory usage and threads vs processes.

A lot of that memory usage will be cached objects, cached Sybase queries, and the Gadfly database, which is kept in memory.

I've been trying to see what effects memory usage and the single biggest factor, beyond the amount and type of "stuff" in you ZODB, appears to be the 'target size' and 'maximum time between accesses' parameters on the Cache Parameters tab of the Control_Panel/Database page. The default of 400 objects and 60 secs between access does keep the memory consumption down, I don't recall off hand how low we where running but around 100M was about right. Recently I jumped these values up to 450,000 objects and 5 minutes between access, we currently have around 300,000 ojects total in the database. This change brought about quite a different memory usage pattern. Once load began to rise we saw one server reach 400+M memory usage before I restared the server w/ a reduced time between accesses value (that machine only has 512M). Our other machines restart too frequently to see just how high memory would grow. -- ------------------------------- tonyr@ep.newtimes.com Director of Web Technology New Times, Inc. -------------------------------