THanks for the useful information herer.
however ,looks like vscsistat only monitor individual vm, not the host.
some more background here. I ran esxtop for two hours with hundreds of VM running. 50% of workload is write op, which cannot be cached I think.
I can see a smooth IOPS curve, so there was IO number growing to peak but no IO sudden spike . this is local disk so there is no other IO tools available . The RAID card has 512 mb cache, but that shouldn't cache so many writes for two hours.
ANy more suggestion to probe into the IO path. Thanks