Quick question around the block sizes that appear in the dashboard. Sometimes when running some load generating script where we use a 4K block size, in the dashboard we see 16K or even at times something like 512B. It doesn't seem to correlate with what we are actually pushing. Is this because of some kind of coalescing going on in the background where it's combining writes? Also, the block size displayed in the GUI, is that what the array is receiving from the host (front end) or is that what's being written to disk (back end)?
Lot of OS / FS has IO elevation. What tool / process you are using for generating synthetic workload?
Also block size displayed is what requested by host.
The workload was generated with some script customer had written.
I've recently spoken to an XtremIO expert and I think I understand what's happening. Seems like there is some host level activity going on at the HBA or multipathing software level that is coalescing the writes or doing some optimization that results in block size discrepancy between what's in the script vs what's on the GUI dashboard.
Also, I learned that the 512B writes are VMware heartbeats (unless you see a ton of them, in that case it could be alignment issues).