On Thu, Jan 10, 2002 at 11:02:21PM +0100, [EMAIL PROTECTED] wrote: > The problem is more complex. You have to provide some details to get answers > that make sense: > > - what is the quota limit of each user?
> [ ... ] In this particular instance it is actually easy for us to come up with these numbers. We have a degenerate case, in which a large number of users are expected to access a mail store which is empty most of the time. quota limit: 100kB people reading mail per day: 100k messages per day: usually zero maximum mail fetch on one connect: 1 expected simultaneous connections: unknown new connects/second: 175 I suppose I could derive the number of simultaneous connections from the connect/second rate if I know how long each connection would last. I imagine they would be very short, on the order of 1 second on average, so the simultaneous connection rate would be 175. I realize these numbers look absurd, but this is not a typical mail application. > A more real life scenario where I did some tests with the following > assumptions (but IMAP, not POP): > > * 3MB quota each user > * we built 80.000 test accounts on two machines (40.000 each) > * we stored random mail in all test accounts (using "postal") > * we simulated approx. 2.400 users logging in distributed randomly over > 1 minute but then reading an average of 3 mails, between each fetch > of a mail 30 seconds of "read time". This resulted in 2.400 simultan- > eous IMAP connections we stretched over 30 minutes as we followed every > disconnect with a new user. > > The setup (2 stock PCs with 2GB RAM and 2xCPU at 1.xGHz with Linux-2.4.x > and ReiserFS) did not glitch very much, load never went beyond 4 I think > and - thats more important - connect/select/fetch times of all clients > almost never did exceed 1 second. Thanks, this is a useful data point. -Guy
