HP’s New Polymorphic Simplicity With 3PAR – The Right Fit For Virtualization?
A large, dimly lit room holds a packed throng teaming with members of the press and a handful of independent bloggers at HP Discover in Frankfurt, Germany. In front of us, a large sheet with 8 HP 3PAR cabinets hides some secret with the words “Effortless tier 1 storage…” emblazed on it. After some momentum and build up has been generated by Dave Donatelli, EVP and GM of the HP Enterprise Group, we are introduced to David Scott, SVP and GM of HP Storage.
Removing the large sheet shows a little 4U box in familiar 3PAR colors. The secret is out: HP now has a mid tier box to offer customers in the form of the HP 3PAR StoreServ 7000 series.
The idea of transformation is a keystone piece of the press release, including the concept of Polymorphic Simplicity. It is a little bit of a silly term if only because it takes something simple and adds a complex word in front of it, but the gist (as explained by David Scott) is that it exists “in several forms, shapes, and sizes” by the delivery method of a single storage architecture for all tiers of business. Sound familiar? It definitely echos of the NetApp approach, in which one operating system, labeled OnTAP, is used for all tiers of arrays. Indeed – a member of the audience did bring up a similar question and was reminded that the hardware configurations for 3PAR are much less limited in terms of scale out (adding additional controllers to the array itself) and scale up (non disruptive migration of data to a larger array class). I will agree that both are important features.
Mixed Virtual Workloads
One of the more key fundamentals I preach for storage that must handle virtualization is the understanding of complex, mixed IO characteristics. I’ve seen many different approaches to handling this, from tiering, caching, using flash drives, and hybrid drives. 3PAR seems to tackle this in a somewhat similar approach of tiering coupled with hardware level thin provisioning. While I’m certainly no expert on how this happens at the hardware layer, the claims are quite impressive. Take this HP 3PAR 7000 Data Optimization Software Suite document for example:
The inclusion of HP 3PAR Adaptive Optimization software in this suite helps deliver an increase in IOPS and reduced latency at a lower storage cost than traditional storage. Adaptive Optimization operates on the principle of I/O access density, which is the observation that for most applications, a very small amount of the total user data receives most of the I/O accesses.
OK – this makes logical sense. The idea that the majority of the data is “at rest” in a typical scenario, and a small quantity of blocks are actively accessed. Continuing …
By configuring only the very active data blocks on Tier 0 storage and the rest of your application data on SAS or nearline SAS storage, an HP 3PAR StoreServ 7000 Storage system configured with a small number of solid state drives (SSDs) can deliver higher performance at a lower cost than traditional storage platforms. HP 3PAR Adaptive Optimization works with the HP 3PAR Security Software Suite to analyze actual user I/O patterns, which enables Adaptive Optimization to move only the blocks with high I/O activity to premium SSD storage.
Sounds like something like EMC FAST technology. Now here is where it gets a bit interesting in terms of controlling QoS for workloads:
… you can prioritize each application by configuring up to three storage tiers (each defined by drive type, RAID level, stripe width, and/or radial placement of data). From this point on, Adaptive Optimization intelligently monitors sub-volume performance and applies the user-specified policies to move data autonomically and nondisruptively using QoS gradients to shift that data at a granular level toward the most appropriate resources according to changing QoS demands. Support for multiple Adaptive Optimization configurations and for the coexistence of tiered and non-tiered application volumes provides the flexibility to consolidate a wide range of applications onto a single array.
This reminds me of what Storage IO Control (SIOC) does for a vSphere datastore at the hypervisor level, but being at the actual block would give it greater control and quicker responses, perhaps? I bring up EMC so often because nearly every slide that HP presented had a direct jab at EMC in some way, so it seems that is who they are addressing as a main competitor for this technology. While I don’t claim to be a storage expert (far from it), it does seem that things are getting more interesting for the consumer – which is always good.
Replication – Perhaps Stretched Clusters?
One thing I need to dig more in to is the option to leverage Peer Persistence for VMware. Through the acquisition of a HP 3PAR Replication Software Suite license, it appears that you can non-disruptively fail arrays for virtual workloads. From the HP document:
This optional suite bundles HP 3PAR Virtual Copy with Remote Copy Software, both also sold separately for all HP 3PAR StoreServ 7000 models. Included in this bundle is Peer Persistence, which enables peer federation for VMware clusters to improve overall availability. Licensed based on system capacity.
Is anyone else reading this and thinking a stretched cluster? Typically, the “hard part” is the storage, as we don’t want a workload running on local compute that is accessing remote storage. Perhaps this may only enable a more robust version of disaster recovery. Still, it sounds very interesting as few solutions are supported for vSMC today (Read Duncan Epping’s article on vSphere Storage Metro Clusters here).
Upgrade To The New Hotness
One additional little tidbit I enjoyed watching was a brief demonstration of the EVA to 3PAR StoreServ software. Per the press conference, you can “migrate online and non disruptively from EVA to 3PAR StoreServ 7000″. If true, it would make getting off your legacy EVA array much less of a hassle, although I would imagine some caveats exist to ensure the workload transitions smoothly. In the case of a vSphere environment, it may be less risky to just fire up both arrays, zone them to the fabric, and issue some storage vMotions. David Scott assured the audience that the migration is “effortless to the customer base” and the StoreServ 7000 is an “EVA on steroids”. Not to be a negative nancy, but the EVA definitely needed some steroids.
No matter if you hate or love 3PAR, the HP announcement underlines the fact that they are making a big move towards a single architecture with plenty of options for expansion, migrations, and enterprise features “for all”. I’ve found that storage is often the most complex piece of the puzzle in virtual workloads, so the idea that you can get more “enterprisey” features in the lower end is great for consumers. Additionally, the idea that you can learn the 3PAR DNA at an early stage in your career (such as on this new 7000 series) that will translate to a 10000 model is a Good Thing. I felt much the same after learning OnTap 7-mode on a 3270.
And, although I’m not a huge fan of the Polymorphic Simplicity motto – at least the idea underneath it (make it simple and available to everyone) seems like a sharp way to play the game. For more details and a chalk talk, I also suggest visiting Calvin Zito’s blog post on the HP 3PAR 7000.