Rather than get into the gritty details, I saw today's announcement as just another step in the journey of storage becoming a fully virtualized and dynamic resource, much in the way networks have been, and servers are fast becoming.
Perhaps no storage technology has been more discussed this year than EMC's FAST. We believe that technologies such as FAST will change the economics of storage in much the same way VMware has changed the economics of servers.
A small amount of data responsible for the majority of I/Os. And almost all of it is infrequently accessed.
The same effect you see on your C: drive plays out in most enterprise applications. Unfortunately, the hot spots tend to move around.If we look at modern storage media, enterprise flash drives are amazingly fast and reliable, but are much more expensive than FC disk. At the other end of the spectrum, SATA drives are amazingly inexpensive, but aren't generally fast enough for most enterprise workloads.
FAST answers the question -- how do you get the best of both worlds?
The answer is conceptually easy: you watch how the data is being accessed, and you dynamically put the small amount of popular data on flash drives, and the vast majority of infrequently accessed data on big, slow SATA drives.
BTW, this is the exact same principle you'll find in CPU onboard caches and other parts of the technology architecture landscape.
FAST technology has been in many customers' hands for evaluation during most of 2009. We wanted a lot of stick time before unleashing it to the general public. As of today, it's ready for prime time.
Simply put, FAST does exactly what we say it does.
* it dramatically improves the performance of the vast majority of I/O intensive applications.
* it dramatically reduces storage costs by increasing the use of low-cost SATA drives
* it does all this without much effort on the part of the storage administrator
* it makes both the business and the IT organization very happy campers
And it does all of this today.
It's the rare foundational technology that nails the value proposition almost 100% of the time, but that's what we have here, folks.
As a result, EMC competitors generally hate FAST with a burning passion, and we will likely see some pretty vehement reactions.
That's to be expected. If you think about it, it's pretty disruptive technology.
These vendors (and perhaps the consultants and partners in their employ) will attempt to claim:(a) it's not here yet [false, it's GA today],
(b) it's not proven in demanding enterprise environments [false, that's where we tested it],
(c) it's too expensive [false, it actually saves big money compared to traditional approaches],
(d) it's too difficult [false, it takes minutes to set up]
(e) or, in some cases, they've got, or will have eventually, a better approach [highly unlikely, but certainly debatable!]
I should point out that these are the same vendors and the same arguments we heard when EMC first made EFDs (enterprise flash drives) available in January of 2008, catching most everyone by surprise.
None of these claims proved to be true back then, and none ofthem will prove to be true now either. For example, remember the vendor FUD on wear-leveling? Sheesh ...More can be found in the extended press release.
Virtualized Storage In The Context Of Virtualized Servers
To follow along in this part of the discussion, it's helpful to temporarily suspend your preconceived notions about the last 10 years of "storage virtualization", and join me in a journey in how storage is starting to journey down the same virtualization path that servers (and networks) have already started down.
The first wave of storage virtualization was all about efficiency -- claw back all the wasted CPU cycles by cramming more workloads on fewer servers.
In the storage world, these are the efficiency technologies: virtual provisioning, data deduplication, and ... FAST. A lot there today, more coming. Basically, show people how to spend less and get more.
The second wave was pooling of servers through cooperating hypervisors. Not only did that further increase efficiencies, it was now much more possible to start dynamically adjusting QoS for different workloads. In VMware land, this was Vmotion, DRS and related technologies -- albeit usually in a single cluster in a single location.
In the storage world, we're seeing the same thing -- consider the V-Max architecture, for example. A pool of scale-out storage resources that not only offer greater efficiencies, but allow dynamic adjustment of QoS for different workloads. Yes, FAST is a key enabler of that concept, but there's more to it than that.
The third wave (where we are now) is the re-orientation of the operational and management model to the virtual machine construct. Simply put, express your policy desires at the VM level, and the rest of the infrastructure follows -- server, network and storage.
In storage land, this is expressed, perhaps oversimply, as "VMware integration". There's far more to it than that, of course. It's the acknolwedgement that the fundamental unit of IT administration has become the virtual machine. Provisioning of storage is done as an extension of provisioning the virtual machine. Specification of services (availability, performance, backup, security, etc.) are done at the VM level, and read by all.
And the fourth wave (early days) is extending these concepts to include geographical seperation -- clusters that intelligently span multiple time zones, for example. All three of the VCE vendors (VMware, Cisco and EMC) are targeting this use case as one of the primary focuses of our joint R+D.
Lots to talk about here during 2010 -- looking forward to it!
But What About Storage All By Itself?
Sure, all of this is great when we're talking about server virtualization, but isn't there a better model for "virtual storage" independent of the host, operating system and hypervisor?
Yes, there is.
And, as part of this announcement, we're publicly sharing just one piece of that expanded "virtual storage" viewpoint. Once we fully abstract logical from physical, many interesting things are possible.
More on this in the next post ...