Mid-tier Storage Accelerates

The bread-and-butter of the storage industry is mid-tier storage -- pragmatic dual-controller designs that do it all for a big piece of the IT landscape.  It goes by many names: mid-tier, modular, unified, etc.  Dual-controller designs provide a nice balance between performance, capacity and availability for so many storage requirements.

Whether it's a modest IT shop, a modest project in a larger shop, or a whole fleet of them in a really big shop – these dual-controller storage designs continue to be a staple of the industry.

It's also a hotly contested portion of the market -- just about everyone has offerings in this space: EMC, NetApp, HDS, HP, IBM, etc. etc.  In a notoriously competitive segment, this is where it can get most competitive.  Fights break out over minor things like the definition of "unified", for example.
 
While this week's VPLEX announcement from EMC World is most definitely cool, it did tend to overshadow a rather compelling chunk of new mid-tier capabilities for  EMC's CLARiiON and Celerra lines.
 
At any other storage vendor, these announcements would officially qualify as Big Stuff.  However, at EMC, they tend to get overshadowed by Bigger Stuff.  I consider that a high-class problem to have :-)
 
See what you think ...

 
To Begin With
 
This can be a tough segment of the storage industry to make a decent living.  Every vendor is on their game, there are plenty of good products, and most product goes through a value-added channel.

As we go through these new capabilities, I'll be doing my best to compare and contrast with some of the other industry players. While doing so will undoubtedly draw their ire, I'll do my best to avoid over-drinking EMC koolaid and keep away from an "awesomesauce" approach.

Thinking About Storage Efficiency

Storage efficiency is important everywhere in the storage marketplace, but here it receives extra focus from both customers and vendors alike.

It's easy to focus on one aspect or another of storage efficiency at the expense of other considerations, so -- as we go through this -- please keep in mind our goal is to advance all aspects of storage efficiency: (1) cost per unit of capacity, (2) cost per unit of performance, and -- perhaps most important (3) flexibility to quickly adapt to new requirements.

The first two aspects are generally understood.  The third one becomes important when life throws you something not planned for -- a new application or workload, a new protection requirement. 

To quote the Boy Scouts -- be prepared.

FAST Sub-LUN
 
EMC has been delivering the first versions of FAST -- fully automated storage tiering -- since 2009.  Our fundamental argument is that this sort of technology has the potential to fundamentally change the economics of storage -- not only the cost model, but the operational model as well.
 
Our first round of FAST has done well, and it's getting good traction.

So, with this announcement, we're making a good thing better -- the CLARiiON (and, inherently, the Celerra) now get sub-LUN tiering.

The chunks are now smaller, which means not only better storage utilitization, but faster reaction time when things change.

Setting up and monitoring this new capability is now dead-simple -- you'll see this during the EMC World keynotes and demos.
 
From a competitive standpoint, some vendors do this sort of finer-grained automovement from flash to FC to SATA, many don't.  From my perspective, it's quickly becoming "table stakes" in the storage world -- much like RAID has been for a while.
 
Bottom line: spend less on storage, deliver higher service levels, and do so with less effort.
 
FAST Cache
 
One of the challenges with any dynamic storage tiering approach is dealing with the unexpected -- e.g. a piece of data is sitting on SATA, and it's gotten really popular all of the sudden.  Not every application has well-behaved performance characteristics :-)
 
That's where storage cache comes in.  One vendor in particular has gone down the path of create large cache cards that sit in front of SATA that can be very helpful in environments where you're re-reading the same data over and over again.

We thought there might be a better approach, and FAST Cache is the result.  The CLARiiON's FAST cache can be used for reads *and* writes: it's non-volatile. 

Instead of building a custom board to do this, we decided use the exact same enterprise flash drives, but just in a different role.

This yields all sorts of benefits -- flash drives can be manually re-purposed between being storage targets and dynamic cache (same physical device), caches can be very large if needed (2 TB of nonvolatile cache on a CLARiiON, anyone?), the data protection techniques we use for drives also apply to the cache, as flash drives come down in price we can pass that on directly to our customers without having to spin a custom board, etc. etc. 
 
Very clever stuff, IMHO.
 
One of the more striking benefits of FAST Cache is that we can now get far more aggressive in our tiering of information downward.  If we turned off write caching, used SATA exclusively, and charged more, it'd look exactly like a certain competitor's offering ...
 
Bottom line: get more aggressive on tiering, create uber-performance environments for specific workloads, and do so for less money --  and with less effort -- than alternative approaches.
 
LUN Compression
 
Looking for primary storage data reduction you can count on??  It's here -- LUN compression is now added to the mix.  And we've done a good job of characterizing its behavior in a number of performance-intensive environments. 
 
Easy to set up, easy to monitor, etc. -- it's just another attribute of a volume.
 
I'm sure the debate will rage on who's got the best approach, which use cases are better, etc. – we’ll leave that exercise for later.
 
And, just to be clear, LUN compression complements (and works with) all the other cost-reducers in the product: thin provisioning, spin-down, etc.

Bottom line: save money on storage capacity with almost no effort, and don't be so afraid of performance-sensitive environments, please. 

Especially in the context of FAST and cache.

 
LUN Encryption
 
No, it's not part of the array, it's done at the HBA level with PowerPath, now with an accelerated option via Emulex, with keys managed by RSA Key Manager.
 
Theoretically speaking, there's a wide range of different approaches to encrypting primary data at rest -- in the application, at the VM level, at the filesystem level, at the HBA, in the switch, in the array, on the tape drive, etc.
 
Hard to declare a winner in the “best approach” category (and we at EMC have to support all of them!), but there's been a strong gravitation to the host-based per-LUN model that we've noticed, since it seems to have a good mix of granularity, control, performance, etc. More to be done in this space, though.
 
Bottom line: nice to have in your toolbox when you need it.
 
UniSphere
 
EMC is currently re-thinking and re-architecting storage management for the coming decade, and you can see some of our first efforts starting to come to market.
 
Viewed tactically, UniSphere provides a clean, unified task-oriented interface for managing block and file storage presentations.  There's an "extended context" of help and support resources nicely integrated in: communities, downloads, etc.
 
Viewed strategically, it's a more compelling picture. 
 
First, there's the notion of a pluggable framework -- both "below" and "above" UniSphere. More and more EMC products will have element managers that can either be used standalone (traditional model), or plugged in so that they integrate in-context with other storage tasks.
 
Going further, UniSphere has an aggressive roadmap to plug into other application and management frameworks, including (naturally) vCenter as a logical starting point.
 
From my perspective, this allows a natural evolution and flexibility in the evolution of storage management models: traditional uncoordinated element management, a more integrated approach if desired (storage and related functions), or environments where certain management functions are performed higher up in the stack if desired.
 
My advice?  Watch this space -- there's some serious new thinking being done here :-)
 
Bottom line: more productive storage management for today's world, foundation for tomorrow.

And, of Course, Even More VMware Integration

Lots of progress in further integration of storage into the VMware environment. 

We've now started to consolidate and unify our various VMware plug-ins across the storage portfolio (e.g. One Plugin To Rule Them All), and have started to incorporate more active management capabilities (e.g. provisioning in context). 

This release takes a big step forward towards these goals.

You may also have heard about VAAI -- storage that intelligently offloads many VMware tasks. 

Although VMware has not formally announced this capability yet, they've talked about it publicly before, and it's coming soon.

With this announcements, we're saying that today's EMC storage products are ready to support these important new capabilities Day 1.

 
Does This Change The Game?
 
Yes and no.
 
In one sense, it's just another round of product enhancements that redefine the table stakes in this dynamic segment of the storage market -- better dynamic optimization, big non-volatile storage caches, new data reduction tools, a few security-related features, a spiffy new management environment, and oodles more VMware integration.

Given the competitive nature of this segment, it certainly raises the standard for everyone who plays here: not only the vendors who are offering products here (they have some work to do!) but from customers who demand the very most from their storage investment.

Who said storage was boring?

Comments

[nate]

what size blocks does sub lun FAST operate in? Myself I'm not yet convinced automagic storage tiering can properly exploit the advantages of SSD. The recent SPC-1 numbers (more specifically the cost numbers) posted by IBM with their easy tier(1GB blocks) reinforces that stance.

the fast cache stuff certainly looks interesting, it's a concept I have been hounding on for a long time now. The key being that the SSDs are used for writes as well as reads. There's more than a couple SSD acceleration systems out there now but this Fast cache is the first one I've heard of that does writes.

[Chuck Hollis]

Hi Nate -- all good. The best blog for this discussion (plus impact on production workloads) would be Chad's blog over at virtualgeek.typepad.com.

The auto-tiering done by FAST, along with the FAST cache, work well together IMHO. The penalty associated with over-aggressive down-tiering appears to be well mitigated by modest amounts of FAST cache. And, as you noticed, the scheme works well for both reads *and* writes.

Add in compression for primary volumes, and it's a pretty good package IMHO.

-- Chuck