If you're talking about Linux, the single most useful tool we used when qual'ing hardware for camera capacity at Intransa was latencytop, followed by iostat. On the windows side of things, perfmon gives you a lot of good stats, queue depth is the most important one. dstat gives a nice overview but doesn't tell you about latency, which is the most important thing when it comes to dropped frames. The big problem with RAID6 is random writes, which require pulling a full stripe frame into memory, modifying it, and writing the changed disk sectors on three drives back out, basically reducing the throughput on random writes to that of one drive. RAID10 works much better in that situation, but of course wastes a lot more space. You may not think that video storage would do a lot of random writes, but the way that most VMS software stores data does have a significant random write component because they are continually appending data to the ends of multiple files.
Another issue a lot of people overlook is switch quality. One reason why Intransa shipped a SMC switch with every storage system was that we found that typical consumer switches simply cannot handle full speed traffic on multiple ports on a continuous basis, they're more for consumer type operation with short bursts of traffic. The end result with a consumer was typically pretty gruesome, the switch would start dropping packets and things... would... get... very... very... slow. Not to mention that under the most severe conditions the resulting dropped connections pollute the Linux network stack skbuf pool until the Linux storage system runs out of memory and either panics or just quits serving data. So make sure you have a good quality switch -- I generally recommend HP Procurve or Dell enterprise switches nowadays, but the latest Netgear "Pro" gear works pretty well too, SMC/Ericcson-LG seems to have pretty much given up selling switches directly in the USA -- or you may have "mysterious" issues even if your storage system is up to snuff.
.
