The last time Omri Palmon from XIV blogged, there was a lot of discussion about 7200 RPM SATA drives and whether or not wide striping across lots of them would provide performance equivalent to legacy storage arrays with more expensive 15k RPM SAS drives. The thing I don't understand is why a vendor wouldn't want customers to have a choice of drives and allow them to use one or the other - or mix them, as 3PAR does.
Omri posted again today on the intangible costs of storage and I had to re-read it several times to figure out if it had been written by 3PAR originally. I don't think so, but I love the way it calls out many of the the benefits of a 3PAR InServ array.
Here's my favorite quote:
"Consider, also, the system’s lead time in responding to user requests. For example, how long does it take the system to resize a volume? Imagine a different reality, in which a storage administrator – or even more junior IT team member – can provision storage at will, in a data-safe, virtualized environment."
In case you weren't aware, 3PAR InServ arrays provides virtual view, role based administration of storage, which means more people can productively manage our systems, without getting in each others way. How, you ask? Wide striping makes it possible.
Just goes to show how end user storage needs are common (universal?).
Unique "how we do it" stories are relatively easy to tell. The real challenge comes in communicating how the end result is uniquely valuable to the customer.
What is 3Par's unique customer-delivered value, in 20 words or less?
Posted by: Pete Steege | February 03, 2009 at 06:02 AM
An interesting read (comments included) but as a customer looking to balance cost, performance AND capacity I can't say I agree with the XIV one size fits all approach. I'm a big fan of tiering and generally matching spindles appropriately with workload. Whether it's done in a single large chassis (like 3Par) or a pool of smaller enclosures (like EqualLogic) really doesn't matter so long as I have the flexibility I need in a single SAN/fabric/management scope (can't quite find the right word there).
Posted by: Chris Fricke | February 03, 2009 at 07:27 AM
Pete, I don't agree with your assertion that there are universal requirements, although I would agree that there are common needs at a macro level, such as wanting to save money on their storage operations.
"How we do it" (architectures and design) discussions help customers get an idea of how they can be applied to meet their specific and unique requirements.
20 words is a very broad brush but here goes....
Enterprise performance and availability that's broadly scalable for all workloads. Simplified administration and autonomic resource utilization reduces CAPEX and OPEX.
Ok Pete, so I'll give you the same opportunity here. 20 words on one of your drives, (Pipeline maybe?)
Posted by: marc farley | February 03, 2009 at 08:56 AM
Chris,
Yes, finding good words for storage "stuff" is often a problem. Given that both 3PAR and EqualLogic have distributed architectures, one problem is defining the boundary of a "system". I guess the word "scheme" could work, but it sounds a bit "snidely". Maybe "implementation"? Maybe "mashup"? I like "distributed system", even if it is a mouth full.
Posted by: marc farley | February 03, 2009 at 09:04 AM
Pete-
It's real simple. 3PAR's unique value can be stated in less than 10 words.
"3PAR allows customers to serve more, with less."
That's it in a nutshell. Much of this ability comes from wide striping, but customers care about the value, not the means to the value.
3PAR can "serve more with less" in many ways.
More capacity with less purchased disk. More provisioning activities with less hours of work by the administrators. More applications using an array with less complication and less wasted space. More aggregate IOPS without a degree in rocket science.
Posted by: Bill | February 03, 2009 at 11:23 AM
Marc,
Mashup is probably the right choice. Though Distributed System makes sense when you consider that another thing folks like me are often doing is implementing different solutions for different workloads/tiers. For example we now have some Data Domain doing awesome things with dedupe for our tier 3 storage which is drawn in the same picture as our tier 1/2 spindles.
I see all of it as "The SAN" but it is a distributed pile of spindles with varying levels of... well simply everything.
When I choose to add a 3Par/Netapp/Xiotech/EMC/Whatever I won't be building a separate SAN but adding to what I've got. The forklift/one-box-to-rule-them-all approach simply doesn't work in the real world anymore. At least not in mine :)
Posted by: Chris Fricke | February 03, 2009 at 05:38 PM