Storagebod Rotating Header Image

Flash, Bang…Whallop!

Like Barry Burke, I found IBM's latest white paper on Enterprise Flash Drives disappointing; as far as I am concerned EFDs are a huge step forward in storage and for the time being ignoring new technologies which may or may not come to fruition in the future; EFDs are a major new weapon in the storage architect's arsenal.

I may disagree with the current fairly crude implementation of them in the EMC arrays but they were first and nothing is perfect first time round. I think it is about time that we stop throwing FUD at the technology and we as an industry start to work out the best uses of them.

We could as EMC have done, 'simply' just create a new LUN. This has advantages in that it is relatively simple to implement but then requires a fair amount of understanding from the end-user to use. I will admit that I fell into the same trap as the authors of the IBM paper in that I immediately knee-jerked and thought putting DB logs onto them would be a good idea. Actually, as Barry points out; not such a great idea.

I must disclose that we had some work done by EMC looking at one of our applications and yes, we could get some good improvements with EFDs but at the cost of adding in complexity to an already horribly complex array layout. And with a growing and changing application, regularly re-laying out the array with all the work that brings could end up a real possibility.

We could as SUN have done look at implementing in the file-system and do clever things there. Disadvantages include you are tied to a particular file-system and potentially a particular operating system if you so wish. But it might be easier and a more efficient use by the end-user.

We could as NetApp, simply…well, simply do something with a RAMSAN device. I've not looked too closely at it; it could end up as complex as the EMC approach. And we also know that they will probably simply allow you to add a tray of EFDs to a FAS; it'll just be a super fast aggregate.

We could use it as a read cache device for workloads which show large random read I/O profiles over a sustained period.

What we do know is that it will make life more complicated and there will be a vendor war about the best way to use them. But perhaps, we need to move on from arguing whether or not that they are good things? And I'll continue to use the term EFD, EMC may have come up with the term but it fits.

I'm interested to see what IBM come up with but I hope it's not more white-papers saying what they might do; I'd like to see what they will do. I'd like to see well-defined EFD utilisation strategies from all of the vendors and then lets see the debate around the strategies.


4 Comments

  1. Geert says:

    Martin,
    I you haven’t done already, take a look at NetApp’s plans on howto integrate SSD/Flash (smartly versus crudely) into our storage systems here: http://media.netapp.com/documents/wp-7061.pdf.
    Cheers
    G

  2. Martin G says:

    Sorry Geert, I don’t see any real smarts there! I can either stick a big read cache in, that’s not rocket science or I can use EFDs as a LUN with NetApps normal features. I don’t see any more smarts in your approach as opposed to the opposition; it’s still crude!
    Yes, I might be able to use the EFDs a little more efficiently due to dedupe. And yes, I could use FlexCache if I’m using NFS but that doesn’t help on CIFS/iSCSI or FC does it?
    Also, once you put trays of EFDs in the FAS, are you not going to rapidly depart from best practice in that when you start building aggregates, all the disks in the aggregate will come from the whole tray? I thought best practice was to try and building aggregates up from disks in separate trays/loops if possible?
    I am hoping that OnTap 8 will be a significant step forward and some of the limitations in OnTap at the moment will go away. I am really hoping that you’ve taken the eighteen months or so to react to EMC’s sledgehammer usage of EFDs to do something clever to drive efficient utilisation of EFDs.
    If you listened to Uncle Joe Tucci today, I think you can see that EMC are going to be refining their sledgehammer somewhat. They know what they did with EFDs, although ground-breaking, is probably rather crude but it allowed them to get them out there and get some real experience as to the best use in one or two real customers.

  3. Yup – Joe let the cat out of the bag. EMC aren’t standing still with EFDs, we’re investing to extend the giant lead we’ve opened up in this new technology.
    Just about everybody are now shipping STEC’s 73GB and 146GB EFDs (everybody except NetApp, it seems). The next step in this important journey can’t be too far off since Joe’s started dropping hints!
    Makes me wonder what kind of dances we’ll be seeing next! ;^)

  4. Martin G says:

    Barry, if we apply the ‘Atmos principle’; we could be waiting for some time! Let’s hope you don’t buy anything, sack anyone or anything else too disruptive which requires moving people around!

Leave a Reply

Your email address will not be published. Required fields are marked *