Author: Ashish Nadkarni, Principal Consultant
Flash drives, also known as solid state drives, have a promising future in the enterprise space. They promise to overcome literally all limitations of traditional hard drives – power consumption, heat dissipation, mean time between failures, speed and IO/s, etc. The list is long. There is no doubt that eventually they will replace all rotational hard drives in the enterprise space, as well as the consumer market. If you have bought into the promise of solid state drives and are planning to invest in it right away, hold your plans for now. The technology should be continually researched to figure out how to best help your environment, stopping short of actually buying anything – yet. The technology is seemingly mature but still has to establish itself in the enterprise space. Then, and only then, will it become viable as a replacement technology for spindle-based drives.
Before we discuss the technical pros and cons of this technology, let us examine the vendor and provider landscape for solid state drives.
It is only a matter of time before current hard disk vendors get into the business of OEM suppliers for flash drives. Today, however, that number is limited to one or two vendors. Keeping in mind the amount of time and money invested in traditional hard drive technology, it is difficult for these companies to do a paradigm shift into a technology that is eventually going to render all of their current investment obsolete.
Among storage vendors EMC is the most notable vendor (and possibly the only one so far) that offers flash drives in their sub-systems. EMC claims that flash drives can form a new “Tier 0” in their storage subsystems, benefiting high-performance applications that demand lots of IOPS with fast response times (see footnote 1). However, in creating this new tier there is bound to be a greater demand for quantification of the benefits that applications existing today on other traditional storage tiers will see. It will be interesting to see if other pieces of the application architecture can catch up or will remain the weak link in the chain, thereby suppressing the benefits of flash drives.
Given that there is not much choice in terms of different types of storage subsystems today the only customers who are in a position to test drive this technology are those that have an existing relationship with, or are planning to jump ship to such vendors. It is not advisable to switch vendors for this reason alone and flash drives do not solve other issues in the storage architecture that may plague day-to-day operations if all options are not adequately considered.
Capacity, Performance and Reliability
Modern Flash drives are made from non-volatile NAND Flash memory. Because there are no moving parts in the drive itself and everything is “solid state”, these drives are very fast compared to their spinning counterparts. Magnetic hard drives suffer from seek latency, which when performing lots of random IOPS can easily become amplified, resulting in data being fetched, written to (or a combination of both) very slowly. Furthermore, latency in a rotational medium is a function of the location of data, hence data response times can vary significantly. Flash drives do not have such limitations and therefore perform very well in such mixed workload environments.
In published documentation EMC claims an IOPS increase of 30 times with response times of less than two milliseconds as compared to 15K RPM Fibre Channel drives. Such performance numbers are unprecedented and if used appropriately will make applications fetch data in a lightning fast manner (1).
All of this performance comes at a fraction of the power consumed by magnetic hard drives. A 64GB flash drive, for example, can use 30 to 40 percent less energy than a 73GB 15K RPM magnetic drive. While this in and of itself may not be significant, when several drives are replaced in a storage subsystem, the savings can quickly add up.
Reduced power consumption also means reduced heat dissipation. Since the drive has nothing that generates heat, the array as a whole will have a lower thermal footprint and reduce air-conditioning requirements.
While they may share the basic architecture with drives found in consumer electronics, the drives for the enterprise are made with better internal checks and electronics that may improve their reliability. EMC, for example, claims that the drives installed in their systems have internal balancing algorithms to ensure that all areas in the drive are accessed uniformly (1).
However, unlike their magnetic counterparts, whose failure and reliability profiles are well known and mature, the reliability profiles of these drives are largely unknown. Not to say that the drives may fail due to component failure, but there is a chance that a bug or a defect in the solid state design or the electronics could pose a challenge to reliability.
Reliability aside, performance and power consumption alone make this a worthwhile proposition until you look at the capacity. In the magnetic hard-drive space, capacity of drives is reaching one TB in some cases. Fibre Channel and Serial Attached SCSI (SAS) drives are in the 500GB range. And then we look at a flash drive and the highest available capacity is still around 72GB. Given that data requirements are not going down, one would require a lot more slot capacity to obtain the usable yield that could be realized by using a lot fewer drives in the former case. There is no doubt that as this technology matures, drive capacities will go up significantly, but for now, there isn’t a lot of capacity per drive.
Today, due to the fact that the primary demand for these drives may lie in the enterprise space, connectivity options may be limited. EMC, for example, houses these drives in a standard Fibre Channel interface. Most vendors may choose to provide Fibre Channel connectivity to such drives as, and when, they support them in their respective subsystems. However, interfaces such as SAS are becoming very popular and it is conceivable that the industry may choose to standardize flash drives on SAS interfaces as opposed to Fibre Channel.
The biggest disappointment in these drives is the cost. Flash drives are not cheap by any standard. What is true for a solid state drive in a MacBook is also applicable to a storage array. The difference is that in the latter case, you may be looking at hundreds of drives and the costs can quickly add up. Again, there is no doubt that costs will come down eventually. But for now, if you want to invest in this technology, better be prepared to write that big check.
So there you have it – another hard-drive technology. However, this one promises to be a lot more important than what it appears to be as of now. In the grand scheme of things, where power consumption of storage devices is coming under scrutiny, rapid adoption of flash drives may provide tangible benefits very quickly. Of course all other moving pieces have to fall into place. – and then there is cost.
Ashish Nadkarni, principal consultant, brings more than 10 years of experience in the systems and storage industry to GlassHouse with vast experience designing and implementing complex storage solutions. Ashish is a regular contributor to Storage magazine’s “Best Practices” column and other industry publications. He has spoken on storage issues at conferences such as The National Association of Information Destruction and Storage World. Before joining GlassHouse, Ashish was a senior systems architect at Genuity; a senior systems engineer at Computer Sciences Corporation; and a training and education consultant at EMC. He holds a Masters degree in Physics from the University of Pune, India and is a member of the Storage Networking Industry Association (SNIA) and the Association of Storage Networking Professionals (ASNP).