EagerZeroedThink! Why Read I/O on Datastore while Inflating

[HMitschele]

Hi All,

i had a call from a Customer 2 weeks ago asking why he has Read Activities on his Datastore when he inflates a vmdk. I did not understand it either and did it in my Infrastructure to justify.

Everytime i created an EagerZeroedThick vmdk, either with the Create Virtual Machine Wizzard (Support Clustering Features etc..) or the context Inflate. As prezeroeing starts 1/3 of the Activity is Read from Datastore and the remaining 2/3 is Write. I expected 100% Write to Datastore since i was thinking that prezero a complete vmdk involves only write Operations (Write zeros). The Datastore does not support vAAI and i have been trying since with EMC and Netapp Subsystems, Fibre Channel and iSCSI. Its allways the Same. With EMC its even more Read than Write Operations.

Does anyone has an Explanation for ths behavior. I looked around and could not find anything.

Thanks

Harry Mitschele

Senior Consultant and VCI

[a_p_ · Accepted solution]

From how I understand it, the zeroing process (not matter if it runs for a newly created or an existing vmdk) only zeroes the blocks that are not already zeroed. Therefore it needs to read each block before eventually writing to it.

Take a look at KB http://kb.vmware.com/kb/1011170 (section "vmkfstools -t0") which deals with the zeroing topic.

André

[HMitschele]

Hi Andre,

that explaines it. Thanks

Harry

[depping]

That is correct, it will read first before it will write the zeroes. By the way if the array has VAAI capabilities it might be worth looking into it to offload this task!

Duncan (VCDX)

Available now on Amazon: vSphere 4.1 HA and DRS technical deepdive