Joel's questions

We have 2 identical (as far as we can tell) apache servers in a load balanced configuration. The servers are behind a hardware load balancer. One of the 2 servers, during periods of relatively high load, routinely exhibits 50% higher process counts than the other server.

The obvious culprit would be a misconfiguration of the load balancer, but the amount of traffic that each server sees is nearly identical (<1% different). The server statistics (cpu utilization) are identical or nearly identical as well.

Apache configurations (the /etc/httpd directory) are identical as well. I suspected a longer tcp timeout on the one server but the thresholds are identical.

What other factors could be contributing?

We have what we believe to be a very common problem but are having trouble coming up with a solid answer. In our production Windows web environment we have 4 redundant public-facing web servers. Content changes are done on an internal-only web server and should be immediately available on the front end servers. Currently we achieve this through CA Arcserve, which seems to work well except for when it doesn't (bugs/crashing/misconfiguration/etc?)

In our unix environment we use a clustered file system (veritas), to achieve this, and are considering switching our windows environment to use the same technology.

What are the common solutions to this problem? Are there writeups of tradeoffs between using clustering technology versus replication technology? What are the most commonly used solutions in other enterprises?