Extents and SAN replication

[kennyfranklin]

I've always operated under the assumption (based on lots of reading and suggestions from others) that volume extents should be avoided at all costs.  Is this still the general consensus in version 4.1?  If you have a server which needs more than 2TB of space, is it stll better to split the disk files between multiple volumes than to create a large extended volume with extents?  If you do use extents, what sort of challenges does it present if you're replicating your SAN volumes?  Has anyone successfully replicated large volumes with extents to a new location and had the volume mount/work properly at the other end?

Thanks,

Kenny

[DSTAVERT]

One alternative to Extents if you need large contuguous space for a VM is to present the storage directly to the OS. iSCSI works very well using the iSCSI initiators available in all OSs.

[kennyfranklin]

Thanks for the reply... unfortuntately this is a fibre channel connection from SAN to hosts... iSCSI isn't an option here (at least not an easy option).

[opbz]

other option might be NPIV but might be flacky or not worth setting up.

For plain ESX unfortunatelly you are limited in VMFS and VMDK file sizes...

By the way option of ISCSI is still valid. I have seen cases where people have presented bigger luns to other (san attached) hosts and then used iscsi to present that lun to particular VMs.. Granted not ideal and messy to set up but did get them over the 2TB limitation

Other options are multiple vmdks and then use software raid on VM... needless to say not ideal

[idle-jam]

the maximum would be 2TB-512kb, so there is a limit the extend would go. we've extended an IBM DS series SAN with replication and it just work as normal. we what did was to stop the replication. extend and create a new replication job.

[kennyfranklin]

Thanks for the feedback.  Out of curiosity, have you actually tried recovering the replicated "extended" volume?  I'm quite confident I can replicate it... I'm more concerned about the viability of the replicated volume since I have to replicate each LUN as a separate replication job.  Re-assembling those LUNs at the other end and having them actually be viable makes me a tad uncomfortable.

[JohnADCO]

Some SAN's I've used have a way to group replicating luns together for this very reason.    But if it's "real time" replication it should be fine, less any potential badwidth issues.  If it's over the wan?   You really should do staged replication.  In that sort of replication you would snap all the quiesced luns before the no I/O's period is over.