5 years on from our purchase of Dell Equallogic, the time has come to refresh our SAN environment. I'm after any tips or features which people are using which I may not know about, to aid in writing the tender specification.
Current environment (feel free to skip to New Environment below if you're not interested!):
~85TB used of EQL group, housing ~50TB of VMFS spread across 16 datastores. Datastores are mostly 2-3 TB, with a couple 5-6TB. One 9.6TB datastore solely used by a VDP appliance.
~40TB in secondary group used purely as a replication target (asynch) for certain volumes. I use Dell VSM to manage that replication.
Of the 85TB in the primary group, it's mostly SAS HDD, except 1 array of 9TB of what EQL call 'RAID6 accelerated' - it has 7 x 400GB SSD in that array.
Aside from the VMFS, we have large LUNS provisioned out to Physical windows servers providing Home and shared drives. These range from 6-12TB.
We also provide some SQL and exchange servers with iSCSI initiators running inside the VM.
Connectivity from Host to VMFS is Active/Active 10GbE using Dell Multipath extension module. VM iSCSI initiators is handled by separate vSwitches connected 10GbE to the separate physical switches to provide diverse routes (ie. each VM has 2 iSCSI vnics, these run the Dell MPIO driver in Windows).
Backups for most 'Tier 1' applications and the large file systems is done by Commvault Simpana to disk staging/tape via an agent in the VM (and physical server for the large file servers). VDP is used to backup 'other' VMs on a daily/weekly/monthly basis and retained for as long as the VDP capacity allows.
ESXi hosts are 8 x Dell Poweredge with 128-256GB RAM in total, plenty of 10GbE.
Performance-wise, I recently ran a 7-day performance trace on the VM Hosts and other volumes on the current EQL, we saw IOPS of ~12000 @peak, ~7500@95% (the VMFS datastore which contains the VDP appliance accounts for around 2200 of these IOPS @95%)
New environment:
Biggest change: we now want to virtualise the large file servers, but rather than have LUNs/Datastores/VMDKs up to 15TB, we are looking at using VVOLS. Does anyone have vendor-nonspecific experiences with these they could share (privately if necessary). I really like the concept of VVOLs - eliminating the slices of 'wasted space' at the VMDK (if thin provisioned) and Datastore level.
Having seen the power of dedupe in the VDP appliance, I'm interested in Dedupe on the primary storage as I think it could potentially give us massive savings especially considering how much OS content could be shared between the VMs. The risk-averse side of me, though, shies away from it. Does anyone have any tales of performance improvement/degradation in real-world scenarios (ie. not marketing whitepapers..!) ?
If we proceed to virtualise the large file servers, I'd like to move to backing these up without guest agents, using Veeam or similar to backup to disk, with the possibility of streaming some backups to tape for archiving/compliance.
I'd like some sort of dynamic tiering of hot and cold data. We have terabytes of archive data which is very infrequently accessed. I'd like the option to force some data to a low tier, but also for the array to let cold data 'sink' over time.
Upgrade-wise, I'd like to be able to do software upgrades on arrays with zero impact on servers. (Isn't that why we buy things with redundant controllers?!) Dell say that EQL upgrades "have minimal impact on initiators" but the full array reboot required at the end (not the controller failover) always causes some fun and games, despite the recommended timeouts etc configured.
I'm not precious about FC/iSCSI, as we're likely to include the network devices for whichever we need in the tender purchase. Although I think if we *didn't* have VVOLS for the large filesystems, I'd still need iSCSI to present to the VMs directly.
If you got to the end of this, thanks, And I'd really appreciate any input.
Thanks
Ian
