Jump to content
Welcome to our new Citrix community!
  • 0

Citrix Hypervisor 8.2CU1 and Dell EMC VPLEX

Emanuele Tonelli


We have a pool of 4 servers connected in HBA to 2 SR (both flash):

- SR1 is on a Dell EMC VPLEX storage cluster (with various Dell EMC XtremIO storage systems connected)

- SR2 is on a Dell EMC XtremIO storage system (without VLPEX in between)


I tried moving the VM between the 2 SR, but the result is always the same: the VM on the SR1 are always much slower (showing many times a disk "highest active time" of 100%).

Any ideas?

Maybe some "write cache" configuration should be changed on the VPLEX?

Link to comment

7 answers to this question

Recommended Posts

  • 0

The storages in the VPLEX cluster are connected in Round Robin.


multipath -ll (of SR1)

3514f0c51d7c00002 dm-2 XtremIO ,XtremApp        
size=12T features='1 queue_if_no_path' hwhandler='0' wp=rw
`-+- policy='queue-length 0' prio=1 status=active


multipath -ll (of SR2)

36000144000000010e019f903b8cf13c6 dm-1 EMC     ,Invista         
size=10T features='2 queue_if_no_path retain_attached_hw_handler' hwhandler='0' wp=rw
`-+- policy='service-time 0' prio=1 status=active


Since Round Robin is the default, this one shouldn't be:

policy='round-robin 0'

instead of:

policy='service-time 0'


Link to comment
  • 0

I don't think this is terribly surprising, you've added another layer of latency to the system in the intermediate VPLEX system so the hypervisor's IO path becomes


Hypervisor -> VPLEX -> XtremIO -> VPLEX -> Hypervisor


and, as the hypervisor always uses DIRECT IO with FORCE UNIT ACCESS (FUA) to ensure that data written is committed to the storage before the IO is confirmed to the VM and thus avoid some of the more common causes of VM data corruption this latency becomes significant. Additionally if the VPLEX adds more latency through processing time of pushing the data to clustering backend then that will be an additional source of slowness over and above that added as a result of the additional interconnect hops.


Whether the VPLEX can be safely configured to complete the response back to the caller before it has completed the IO to the backend storage tiers would be a question for Dell but it would require a significant amount of non-volatile or battery backed memory to be even remotely safe and even then there are risks associated with the backend tiers going down/losing contact and the VPLEX being unable to flush.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Create New...