Solid State Disk (SSD) 

By Brian Feller

"I do not fear computers. I fear the lack of them."  Isaac Asimov

For much of the 19th and 20th centuries, the most critical element to the success of any business was a simple equation: Location. Location. Location.  Businesses lived or died based on their ability to select an optimal geographical area. In the first century of corporate Darwinism, those enterprises that controlled the highly trafficked areas with excellent lines of site attracted net-new clients and survived.  Those that did not went the way of the Tucker, Wachovia and Commodore.

With the flipping of the centurial calendar, this underlying ingredient of achievement has segued from one of physical space to a much more binary algorithm:  Data.  Data. Data.  He who has it (faster, broader) wins. Period. The magnitude of today’s data is growing at a scale that would make even Mr. Moore and his law gasp – yet is arguably exceeded by the enterprise’s need to access it anywhere, anytime.  However, similar to how the U.S.’ highway infrastructure was built to accommodate about 1/20th the amount of daily traffic it actually supports, this has created a major performance gap that has seen hard drive-based storage systems lag behind advancements in Random Access Memory (RAM) and processors.

While this technological revolution has provided consumers and businesses alike with devices previously relegated to episodes of Star Trek (Next Generation-style), it has also created a discernable performance gap. Hard drive-based storage systems have failed to keep up with Intel and AMD as they have released their modern processors. As this gap continues to widen, the concept of solid-state as a primary storage alternative has gained significant steam – being viewed by many storage experts and analysts as the inevitable next phase of Tier One storage. With any advancement in technology comes the corresponding obsolescence of incumbent solutions, thus the battle lines are clearly drawn: SSD vs. HDD.

Slow-Performing Disk

With this dramatic development pace from the major chipmakers, today’s enterprise servers are simply being starved for data. Processing speeds have increased exponentially, while the speed at which a disk spins has not changed in 10 years since Seagate released the 15K RPM drive. This performance delta between the two is only being exacerbated by each release of their respective chips. The underlying issue is this: database workloads are extremely demanding on traditional spinning disk systems. The majority of the traffic is based on extremely small request sizes and is often random in nature. The mechanical latency is a limiting factor in database performance. Barring a cache “hit” (data already in a RAM buffer) each request will likely require the movement of the read/write head incurring a 6–9 ms penalty while the head is moved and the data is retrieved. Adding in the mechanical wear, tear and speed limitations of traditional disks being 15,000 RPM, they can perform at a sustained 200 I/O operations per second; enterprises have been forced to bloat their storage spend to gain even a moderate performance increase. 

Even under the best of circumstances, within an ideal operating environment of an air conditioned office or data center, HDDs deteriorate over time simply from normal wear and tear on the moving recording mechanisms each time they are turned on and read from or write to the disk. The frequency of drive failures, crashed drives and lost data are well documented. Every company has their personal Friday The 13th horror story abound about how valuable data was lost and the anguish of trying to retrieve a portion of a crashed hard drive. Entire industries and submarkets have arisen to help safeguard business entities from the all too common problems with HDDs offering backup, archival, off-site storage, and data-recovery services.  With solid-state technology maturing and making major inroads into the datacenter, companies are now re-evaluating their risk tolerance for hard-disk drives tied to mission critical data.

With the macro-economic headwinds blowing strongly for the past few years, C-level executives are looking for ways to limit spending.  A major target has been the allocation of IT budgets for storage – an estimated 60-70% of the entire budget.   

Some of the high-profile, high I/O use cases where the advantages of solid-state based storage can be realized most quickly are virtualization, messaging, SQL, Oracle and OLTP.

With virtualization being first and foremost on everyone’s minds, let’s put a laser-focus on how SSDs can alleviate much of the pain when related to a storage infrastructure.

Solid State Leverages Virtualization

Virtualization environments are typically over-subscribed when it comes to disk performance. Everyone is competing for the same pool of resources. This is typically not a problem in an environment where low utilization servers have been virtualized. However, when mid-large organizations try to virtualize Tier One applications, major disk performance issues often emerge. Oversubscribed disk systems are notoriously hard to identify. CPU and Memory usage is much more “in the face” of administrators. Disk systems require intimate knowledge of the underlying disk infrastructure and the workloads running on them. Virtualization can make this even more challenging, as most monitoring and management tools only show the amount of data being transferred back and forth.

Since many systems are competing for the same disk resource, it can turn a SEQUENTIAL workload into a much more demanding RANDOM workload (RANDOM being the worst possible workload for a mechanical disk). This is because, in virtualization, every VM gets a timeslice of every available resource (CPU, MEMORY, DISK, other I/O, etc). Thus, a nice sequential file transfer gets interrupted every so many milliseconds by another fileserver or an Exchange server, causing the read/write head to be moved to another track, incurring a seek penalty.  With a solid-state based SAN, average data access time is sub-millisecond (0.1ms), making SEQUENTIAL and RANDOM workloads no more of a challenge than the other.

While virtualization overhead can be seen as low (6–15%), it is still significant and impacts performance for I/O-intensive applications. Solids-state SANs allow enterprises to finally recognize the value of virtualizing this tier of applications by making the “performance pool” dramatically larger, eliminating the contention. Major virtualization vendors such as Citrix, VMware, Quest and Microsoft have all been spending more time with SSD vendors to evaluate their strategies. 

So, where does that leave us?  Clearly we have reached a major crossroads in how we design a storage infrastructure.  Performance, not capacity, needs to be the primary focus for an enterprise’s mission-critical applications.  Hard disk drives are not going to go away. They will simply be handing in their membership card to Tier One and throttling back onto the Seniors Tour, leaving Tier One to the world of solid-state. 

It’s simply a matter of physics and time. 

Brian Feller is VP of operations for WhipTail Tech.