One of the coolest features of the NetScaler switching product is Surge Queue. The idea behind it is simple: don't let more than n outstanding requests be sent to a server at a given moment. As the server responds to a request (creating n-1 outstanding requests), dispatch a new request. If the request rate exceeds the server's turnover rate, queue them. This allows a server to run at peak capacity without the risk of having it spiral out of control because of a surge of incoming requests. You can track the queue length through an SNMP OID. Stir with MRTG and you can visually see the queue length -- if you always having something in the queue, you know it's time to buy more servers. If you only get a periodic surge, then you can relax and let Surge Queue do its thing.

I really could use Surge Queue on me. I think my brain's request queue overflowed yesterday and I'm having to sleep it off today. Ugh.