The Question and Answer Board

[Bob Zumbrunnen]

You're correct. It's crashing every day and whether coincidental or not, it's happened within an hour of the same time each of the past 3 days.

Doesn't appear to be a memory leak. The machine always reports more than enough to spare. 1.7 gig available physical memory right now, and it usually stays around that number.

Very frustrating that the event log records nothing about it. Nada. Zilch.

I've been keeping an eye on the number of concurrent connections and CPU utilization at the same time. What's interesting is that I have the connection timeout set to 9000 seconds (almost 3 hours), so the number shouldn't fluctuate much. However, immediately prior to or coinciding with IIS coming down, the connection count quickly doubles. I haven't been watching the instant a crash happens to know which is the chicken and which is the egg.

In looking at the logs around the time of yesterday's crash, the only thing noteworthy was that there was an absence of iHub traffic right before the crash. Nothing but SI traffic for some remaining ads and graphics, which have since been moved elsewhere.

And since finally getting *all* the ad-handling and some image-handling moved to the ad server, it, too, has started having problems, though utilization remains in the 30% area.

It's really like there's some connection threshhold we're exceeding that we're not aware of existing. I've got IIS set to allow unlimited connections.

Any ideas?

For what it's worth, concurrent connections are typically in the 2-3000 range, which I suspect is a completely bogus number. At some point in using our previous webserver, the reported numbers went from about 14k connections down to 2k suddenly, which makes me suspect a bug in Performance Monitor, but the same low numbers appear on the new machine with a fresh W2K install. With the amount of daily traffic the site gets and the number of unique users per month, it's very difficult to believe that there are only 2-3k people who have used it in any 9000-second timespan during the trading day.

Edit: When the crash happens, CPU utilization drops to zero while the connection count skyrockets, and stopping/restarting IIS doesn't fix it. Only a reboot of the machine does. And Dell's with SCSI RAID setups don't boot particularly quickly.