Friday, July 29, 2011

Where does this performance demand come from? After all, aren’t all the users accessing remotely, typically over a public internet connection? It’s true, they are, but all the processing is done locally, just the queries and the answers are sent across the internet. More importantly, those requests being made simultaneously by thousands - if not hundreds of thousands - of users at the same time.

These environments are also heavily virtualized, with Exchange and SQL databases. This means high levels of random I/O. Typically, a block-based storage system is preferred. The problem is though that most cloud storage solutions being considered today tend to have a NAS front end only.

Pick your favorite social media site. You query for someone, there is a database on premise at the cloud providers data center for millions. maybe billions of contacts that is searched, and just a very small fraction is sent to you, seemingly instantly. Whether it is a social media example or an online business application like CRM these sites are under pressure to quickly handle requests from thousands of users in simultaneous fashion. If it takes too long to get an answer, you will stop using the site. Switching in our online world is relatively easy, so pressure to remain competitive is high.

At the same time, the companies that provide these types of services still need to be price competitive to be able to meet customer price expectations (often free). The option of simply throwing expensive hardware from legacy enterprise hardware vendors simply  is not sustainable for the long term. Providers need a new option and SolidFire is one worth considering.

SolidFire is a solid-state only, clustered scale out storage solution. Unlike other scale out storage systems though, this is a block-based solution, not a NAS, making it ideal for database-intensive environments or environments where an application does not support its data being NAS hosted. Each node in the cluster is made out of industry standard hardware, which contains the solid-state storage and 10GbE connectivity. Each node in the cluster is able to deliver 50,000 IOPS and the total cluster can deliver 5 million IOPS. Performance, as you would expect in a clustered storage system, can scale linearly.

Performance can also be tuned through a unique quality of service technique, which enables cloud providers to prescribe and guarantee both IOPS and bandwidth values to every volume within the system. We expect providers to leverage this capability by creating different performance tiers and charging accordingly. This keeps customer expectations in check while ensuring that performance resources are available for those customers that are paying for higher performance levels.

Other than the obvious performance potential of a pure solid-state storage cluster, SolidFire is also committed to delivering that performance at the same price point as the traditional SAN. They accomplish this by making sure storage utilization is very high, the goal being 85%. They also leverage the latest in space efficient technologies like thin provisioning, deduplication and compression; but do so in real time, thereby reducing the amount of data stored within the system and dramatically reducing capacity requirements. Each node in the cluster has a relative capacity of about 12 TB. While the overhead associated with these technologies may seem to be counter-intuitive to a pure solid-state system, as we have seen with other pure solid-state storage systems SolidFire has performance to spare and the net result is a very high-performance system at a very price competitive rate.

SolidFire also brings a unique cloud management interface to their storage system. Instead of a proprietary management GUI that is difficult to customize, the SolidFire systems can be managed entirely through a REST API. This allows the cloud provider to customize and fully automate management and provisioning to their specific needs.

SolidFire’s systems have a complete complement of data services features as you would expect in an enterprise storage system like snapshots, clones and point in time backup. Remote replication, which will be available in early 2012, will give them a disaster recovery function as well. Protection from SSD failure is done by one or two way replication throughout the nodes on the cluster. This improves performance since RAID is not calculated on a per node level. While it may give you pause to think of clustered RAID 10 on an all-SSD storage system; it is the combination of real time thin provisioning, deduplication and compression that not only make this plausible, but makes it preferable, since the performance and extra write cycles of a RAID engine are eliminated.

Storage Swiss Take

Cloud compute needs a new option. The legacy storage models are just not a good fit for these highly dynamic, highly random I/O environments. Many cloud providers have tried to build their own solutions, but most would rather not be storage vendors, so they turn to the legacy enterprise vendors as they grow. SolidFire may be their way out; cost- effective, high performance solid-state storage clusters built specifically for the cloud compute environment. Keep an eye on SolidFire; we think there is a winner here.

George Crump, Senior Analyst