This is the question that gets asked often, and it’s very heated topic to boot. There are plenty of folks in the industry who says it’s going to stay and plenty of others who thinks it’s on the way out. Google Infrastructure VP, Eric Brewer recently argued: “current disks are designed for enterprise service, not for their current dominant use as a cloud-based storage.”
Brewer explanation about his claim made a lot of sense. I’ll quickly summarize and add some background information.
“SSD” is the short form of Solid State Drive.
SSD technology primarily uses electronic interfaces that are compatible with traditional block input/ output (I/O) hard disk drives. Unlike traditional spinning hard drives SSDs have no moving components.
SSDs are more resistant to physical shock, run silently, have lower access time, and less latency. However, the price of SSDs are still roughly four times more expensive per unit of storage than HDDs.
But what does Google like a massive buyer of hard drives want? They want a large volume of storage as they need it for their mammoth size data centers. Google and other data center service provider are ready to sacrifice reliability over cheap price and large size.
Fast boot and app start is the reason why millions of SSDs are bought every year. But that doesn`t mean SSDs run smoothly and doesn`t cause the problem.
A Large Scale Study of Flash Memory Failures in the Field by folks of Facebook and Carnegie Mellon University highlights some of the operational problems.
Facebook was among the early adopters of SSDs, their experience is more insightful than most. Here`s what they found in a nutshell:
- SSDs are sensitive to temperature more than HDDs. When they hit the hot spot, SSDs may slow down.
- SSDs are power hungry. The team found that as power consumption rose, so did failure rates.
- Level of system write activity is directly correlated to SSD failure.
- The team found 4.2 to 34.1 percent of the SSDs reported a un-correctable error.
Respect to the CMU/ FB team for an important paper. This paper offers an evidence-based view on SSD behaviour that is available nowhere else.
Another research by folks at SanDisk investigates the impacts to performance in a flash when multiple layers of log-structured applications and file systems are layered on top of log-structured flash devices.
We show that multiple log layers affects sequentiality and increases write pressure to flash devices through randomization of workloads, unaligned segment sizes, and uncoordinated multi log garbage collection.
In a nutshell, if we believe SanDisk folks the log structured I/O management built into SSDs are seriously not up to the highest standard and that’s true for all the apps that use log structured I/O (which today most of them are).
In servers hosting virtual machines, it is believed that SSDs are a lot better than disks. But a recent research in Korea revealed they`re far from perfect.
Jaeho Kim and Donghee Lee of the University of Seoul and Sam H. Noh of Hongik University points out:
We show through empirical evaluation that performance SLOs cannot be satisfied with current commercial SSDs.
This paper shows, using SSD poorly can waste most of its possible performance. And without the right controls- the ability to pause garbage collection would be useful- SSDs will inevitably fail to provide the performance they were expected.
SSD technologies are still at their young age. And we are continuously learning about their behaviour in optimum condition.
Since the introduction of SSDs, we have come far away in terms of technology advancement and their usage as well. Despite the recent research pointing out the pathologies of SSDs, we can`t ignore the real purpose of introducing SSDs- Reliability and fast access.
Storage isn’t getting faster, and making them faster is a great help. We can focus more on that along with lowering SSDs prices.
Readers are urged to read the following papers for more details.
- A Large Scale Study of Flash Memory Failures in the Field
- Don`t stack your Log on my Log
- Towards SLO Complying SSDs Through OPS Isolation
Are you surprised or you were expecting the findings? Be sure to shout out in the Comments Below.