Scaling in Windows Azure for IO Performance -
windows azure advertises 3 types of io performance levels:
- extra small : low
- small: moderate
- medium , above: high
so, if have io bound application (rather cpu or memory bound) , need @ least 6 cpus process work load - better io performance 12-15 smalls, 6 smalls, or 3 mediums?
i'm sure varies based on applications - there easy way go testing this? there numbers give better picture of how of io performance increase move large instance roles?
it seems io performance smaller roles equivalent larger ones, ones throttled down first if overall load becomes great. sound right?
windows azure compute sizes offer approx. 100mbps per core. small instances lower, @ 5mbps. see this blog post more details. if you're io-bound, 6-small setup going offer far greater bandwidth 12 extra-smalls.
when talk processing workload, working off queue? if so, multiple worker roles, each being small instance, each work 100mbps pipe. you'd have benchmarking determine if 3 mediums gives enough of performance boost justify larger vm size, knowing when workload down, "idle" cost footprint per hour 2 cores (medium, $0.24) vs 1 (small, $0.12).
Comments
Post a Comment