The Truth about IOPS
I often find myself wondering why so many people are still yakking about these crazy little things called IOPS (input/output operations per second). Sure, I understand that the Solid State Drive (SSD) vendors needed a unique way to market their product and make it obvious to the consumer that this isn’t your typical hard drive. But these artificial numbers supplied by the manufacturer are always describing a pristine situation that you, I, and the rest of the world will never experience! So what gives, and what does all this mean for Desktop Cloud Computing solutions?
Since the advent of computers, accessing long-term storage has always been a bottleneck. And IOPS do matter when you are leveraging mechanical hard drives that can only offer a modest 125 random operations per second. But what we find is that the real-world solid-state technology easily yields 100,000+ IOPS, relieving the bottleneck completely. In fact, these drives are typically waiting for work while the Operating System is frantically trying to issue requests for more data.
So, is any of this news to the technology world? This shouldn’t be anything close to a surprise, as engineers have been coding around the IOP bottleneck since its inception.
In fact, engineers have been busy working on compensating technologies including:
- Readahead, where the OS prefetches data around a given location hoping that subsequent reads will happen directly from RAM.
- Complicated page cache architectures so the OS is never re-reading data that was previously read and stored into the cache.
- IO-elevator algorithms that intelligently rearrange the order of reads and writes to increase effective bandwidth.
We at V3 Systems took the initiative and realized that an optimal customer experience was critical to securing the adoption of Desktop Cloud Computing solutions. By placing the OS on local solid-state storage, we can guarantee that the user will never be impatiently waiting for the storage below. V3’s solution easily provides IOPS that substantially exceed traditional spinning media, almost as if the virtual machine had its own private drive, even under the heaviest of loads. And, when all the virtual machines aren’t starving for data, it’s safe to say that any virtual machine can gobble up to 20X the IOPS of a traditional SATA drive.
Sounds great right? Then why do traditional VDI architects recommend allocating only 30 IOPS per desktop, and benchmarking suites only require 10-15 IOPS for a passing grade? If you ask me, the bar is set so low that even a casual user will be left longing for the clunking loud machine that used to sit on their floor.
I say, “Let VDI evolve into Desktop Cloud Computing solutions! And say YES, virtual desktops need to have low-latency storage architectures to guarantee an optimal user experience!”