Setting the scene
Im one of the many who have been known to beat their drums over the way the storage industry has abandoned the small, fast, disk drive and left it to die a slow death. While on the other hand has killed the fatted calf and lined the streets with palm leaves for bigger fatter slower drives. Especially those that are members of the ATA family.
Databases that would once have lived, and prospered, on their own fast dedicated spindles are now commonly found existing in the equivalent of a large communal dorm with shared kitchen and washing facilities (i.e. sharing the IOPs, throughput and other resources of larger and larger disk drives with lower class ‘citizens’ such as photos and mp3 files). What a crying shame…. almost brings me to tears ;-)
The imminent demise of the disk
For quite some time now there has been very little interest and virtually zero investment in making fast disk drives – even from the vendors who bang on about being performance centric! How many of them even support never mind sell anything 73GB and smaller? Yet Im willing to bet that they all sport the latest chubby SATA drives.
Of course the drive to increase capacity, in and of itself, is not a bad thing. But the way in which it has been pursued is where I have the issue. It has been at the expense of all else, including reliability and rebuild times – lets be brutally honest, SATA disks are manufactured to a lower standard (although this is less so these days) and the associated elongated rebuild times expose you and your data for far longer. Talk about going against the grain of fundamental principals of enterprise storage! Personally I don’t blame HDS for being reluctant to allow their customers to pollute their USPs with them.
Seeing the light
But now, I see why capacity has received all the attention. The vendors and manufacturers were all obviously aware that the disk drives days occupying the tier 1 top spot were numbered. A new Tier 1 (some are calling it Tier 0) was about to be born and the disk drive was destined to slip down the rankings to a spot where capacity was more important than speed.
So the writing has been on the wall for a while, the only uncertainty being who would replace the disk at the top. Personally I think we now have the answer and I think history may well show that EMC’s recent announcement regarding supporting flash based SSD in the Symmetrix will prove to be the anointing of flash based SSD as our new Tier 1 and king of performance. And I have to take my hat off to EMC for being the first major vendor to do it. See, the sun even shines on the baboons ass from time to time ;-)
As I said in answer to a recent question on this topic on the HDS forums, what one top storage vendor does, the others eventually end up doing the same. So I expect the other serious vendor
s to follow suit shortly. There are only two serious vendors ….. Hitachi and EMC right?
So what about capacity for SSD
So if the writing has been on the wall for quite a while, why has it not happened before now? One of the reasons why flash based SSD has taken so long to be recognised as the next in line has been its relatively small capacity in relation to the spinning disk. We often read statements like “..the capacity gap between SSD and conventional spinning disk is closing fast” and I for one was chomping at the bit for that gap to be closed as quickly as possible. But now that the gap is narrow enough to make them viable, Im starting to revise line of thinking on this. I now hope the gap doesn’t close all that fast any more.
Learning from others
I for one will be crossing my fingers that the SSD manufacturers won’t follow, or be forced down the same path as the disk manufacturers. But in saying that, I am not saying that the pursuit of larger and larger capacities is bad, just as long as everything else is not sacrificed on the alter of capacity. After all, SSD is poised to take up its place at the top table of the storage world on the back of one major factor – speed!
Sure it brings a few other great things with it such as lower power consumption and better reliability (yes you read that right), and these are all great, but they are all side issues compared with the performance benefits it brings.
What does it mean for the rest
Amid this I wonder what it spells for the existing SSD based array vendors such as TMS?
Does this validate what they have been saying and doing all along?
Does this make them and their tried and tested technologies more of a serious player?
After all, EMC are new to this and will no doubt stub their toes and graze their knees along the way as they do the right thing and clear the path for the rest who will no doubt follow behind (hey, we all live in the same world and we have a duty to make the way easier for the others that will travel the same path at a later time. EMC have certainly done the following over certain technologies in the past).
Despite the existence of the TMS RamSAN and other SSD based arrays, I believe that EMC are marking new territory with this one. After all, they are putting SSD into a real melting pot of storage, the DMX, which is very different to the TMS kit. They will be the first, that I know of, to shove SSD into the same box as potentially hundreds of chubby 1TB SATA beasts and ask these two performance extremes to live together in harmony while sharing the same physical architecture (ports, paths, cache……).
Im not saying it cant work, and I have to admit as a person who spends probably too much time in the company of monolithic storage arrays that I will be watching with interest how this pans out. One thing is for certain though, it sure as hell makes the DMX the jack of all trades and fits EMCs idea of Tiering in a Box. Next thing we know, they will be adding support for tape in the DMX-4 (joke).
What does it mean for enterprise storage arrays
I used to think that enterprise storage arrays were where you put your servers that required the best response times etc. However, Im seeing more and more and more and more…..enterprise storage arrays that are running like dogs, and not the greyhound type. And every time its for the same reason – they are being used for everything from high end OLTP databases, through messy Exchange servers and ad-hoc development systems all the way down to users home directories. Oh and no consideration is given to performance when new servers are attached and often all RAID groups are pre-cut as RAID5 for ‘convenience’…….. and that’s all without SATA in the mix.
So, is the writing on the wall for enterprise storage arrays to be relegated to jack of all trades masters of none, your general purpose array that you just throw 'most' stuff on. And if you have particular performance requirements you build it on a specialised array that is SSD only? So….. may be there is life in TMS yet.
In conclusion, now that the writing is on the wall, if you don’t know your NANDs from your NORs, your erase blocks from your physical blocks, and more importantly how to make best use of such characteristics, I would suggest now be a good time to start skilling up. Otherwise you might find yourself relegated, alongside the disk drive, to tier 2 of the storage world. Seriously! I will be swatting up!!