SSD and storage efficiency

[ivisinfo]

There is a lot of discussion on SSD drives but only in regards to performance. What about storage efficiency, i.e. the raw capacity required in a real environment?

If I understood correctly, SSD drives will be introduced as normal disk drives. Does this mean that there will be a separate SSD aggregate, together with the separate SATA, FC, and SAS aggregates? What RAID level will be used for the SSD aggregate?

Also, WAFL keeps live (tier-0) data and snapshots (tier-n data) on the same type of disks which is not storage efficient. Apart from the great benefits of NetApp snapshots, in regards of efficiency the technique of keeping snapshots (data that may be never used) on expensive disks introduces a cost penalty. Yes, snapshots are small in size due to the WAFL architecture, but in a real life scenario, snapshots can reach 50% of total storage. And 50% more SSD drives is a big cost.

Does this mean that expensive SSD space will be wasted for keeping snapshots?

Thank you.

[scottgelb]

Correct that SSD are a separate aggregate from other drive types.  Snapshots of flexvols in the ssd aggr will take space on the ssd aggr.

Here are my notes on usable for the SSD drives.  There are 24x 100GB SSD per DS4243 shelf on their own stack.  With more than 1 shelf of SSD, 26D+2P RAID Groups are supported and realistic to use 28 drive raid gruoups.

SSD right sized for a single shelf of 24 drives.

used (MB/blks)        Phys (MB/blcks)
---------------        ----------------
84574/173208064        84796/173663616

1.94TiB raw per shelf   24*84574 MiB = 1.94TiB
24 drives - 2 parity - 1 spare = 21 data*84574 = 1.69TiB
Aggregate capacity net of WAFL reserve = 1.52TiB  (no aggr or vol reserve)

[thomas_glodde]

Why should i dispose the precious snapshot data to lower (slower) tier storage? eg. if i move the snapshot part of my important sap database to slow sata disks, restore takes ages again. the strenght of netapp & its snapshot capacity lies within the fast (we are talking about minutes) restore of production critical data.

if you definately need tiered storage for eg. file service you can always snapvault less important stuff to secondary storage.

netapp is a software company which sells quick and SECURE (!) enterprise storage software products. to keep the product neat & simple, netapp sells the correct hardware for its software products itself. security always comes at a certain price, so if you need secure, high available space, then of course its more then just a bunch of disks.

[sshapiro]

Thomas,

Data ONTAP snapshots are space-efficient and quick to take and recover from because they are simply pointers to blocks that already exist on the storage media being used.  Moving the snapshot data to a different tier of storage - which is in effect what SnapVault does - cannot be as fast or space-efficient.  You can have the data in 2 different places yet share it and recover from one copy to the other instantly - that just doesn't work.

And let's suppose you had a storage pool that consisted of 2 types of storage media - one fast and the other slow. And as a block that was originally written to the fast media aged, and at some point was only referenced in snapshots and not in the active filesystem, you wanted it auto-migrated to the slow media.  That could be done, but of course involves copying blocks from the fast to the slow media, and in the process consuming CPU and drive IO on both the source and destination.  Is it worth the cost?

If blocks are being updated at a high frequency, then you would want 'old' snapshot-only blocks moved...but then this also means a lot of block move workload.  If blocks are not changing much, and thus remain in the active filesystem for long periods of time, then you'd probably want to keep them on the original media.  So when you want to move blocks it's costly to do so, and when it's less costly there's little incentive to move them.  See the conundrum?