Finding The Sweet Spot For High Performance Virtualised Environments

Virtualised Environments

Despite the widespread interest in flash, there continues to be a surprising amount of misinformation and misunderstanding about the technology and its application. Is flash good for all workloads? Can flash be added to existing disk-based storage and produce the same effect as all-flash storage? What are the criteria to use when evaluating different flash solutions, and how can we take flash to the next level in a virtualised environment?

Flash-First Storage Hits The Sweet Spot

Modern workloads are IOPS hungry. Further, storage performance woes are no longer visible to just IT. Take VDI as an example. Any performance issues caused by storage affect end-users. Unless the industry continues to innovate, CIOs will be forced to buy more and more disks just to meet performance needs.

In order to improve performance, some vendors began adding flash to traditional storage as a cache in front of the hard disk. Although the addition of flash will improves performance to some extent, this type of approach won’t provide the same cost and performance advantages of a Flash-first hybrid architecture, which has emerged as the sweet spot for modern mainstream workloads, especially virtualised workloads.

The flash-first approach delivers far superior performance than traditional storage with flash as a cache. They are far less expensive than all-flash systems as well as traditional storage as the need for IOPS grows. Flash-first architectures, with 99% flash hit rates, significantly outperform generic flash-as-cache storage, which achieve only 30-80% flash hit rates.

One of the most interesting findings from analysis of the advantages of flash-based storage is the fact that there is a point at which a flash based system begins to outshine traditional storage arrays from a cost/IOPS perspective.

With 10 TB of usable capacity, at the 7,000 IOPS mark, a flash system and a traditional disk system have about the same price from a performance perspective. Beyond that mark, the cost of the traditional disk system continues to escalate as spindles are added to continue growing performance while the flash system continues to have plenty of available IOPS capacity to meet continuing needs.

Putting it more succinctly, an environment requiring 15,000 IOPS from 1 terabytes of usable storage would require:

  • 64 drives and 1 TB of flash in a traditional storage array with flash as a cache
  • 16 drives including 2.4TB of flash in a Flash-first storage array
  • The traditional storage array would cost more than twice as much as the Flash-first ($190,000 vs. $88,000)!

Response Time

The cost savings is not the only area that flash trumps disk. Another visible advantage of the flash-first approach is the improvement in IO response time. The response time of the traditional array would be three times higher than the Flash-first storage array (1.7ms vs. 5.1ms).

For a Flash-first system, the response time varied between 1ms at 1,000 IOPS to 2ms at 20,000 IOPS. For traditional storage without a cache of flash storage, the response time was about 9ms, and with a flash cache it varied from 4ms to 6ms. This is especially important for workloads with databases, and workloads that have shifting peaks of high IO activity, e.g., VDI, Citrix, or VMware implementations that almost always have IO storms at various times of the day.

Management

Latency-sensitive workloads such as databases can be implemented more efficiently if the IO latency is reduced to sub 2 ms. Low latency storage also significantly reduces the effort of DBAs and system administrators to manage the systems. Users of Flash-first storage can spend fewer than five hours a week in administration, compared to two-to-four times as much for traditional arrays with flash as a cache.

So let’s say you are convinced that flash is the right way to go. How do you pick the right solution? It’s possible to define the primary characteristics of a Flash-first storage solution as:

  • All reads and writes can be serviced from flash. This ensures that the solution will scale with future improvements in flash technology.
  • The storage has been designed from the ground up to use flash as the primary storage media. A disk-based system that has been retrofitted for flash or designed to use flash as a cache rather than primary storage, will never be able to take full advantage of flash.
  • Provides intelligent application-aware management capabilities made possible by flash. Anyone can use flash as a faster disk, but what new capabilities does flash enable that was not possible with disk?

All of these characteristics are important to take full advantage of flash in virtualised environments. They lead to systems that provides greatly simplified management, lower cost, higher performance and greater agility.

Flash Enables Application-Aware Smart Storage

So what is it like to use application-aware smart storage? Remember what it was like to do trip planning before smart phones? You would have to do exhaustive research before your trip and print all materials such as maps ahead of time.

In contrast, with a smartphone, you just need to note the major points of attraction and the smart phone provides you with the information you need when you need it during your trip. Furthermore, the phone has smart apps which automate common tasks such as figuring out how to get from one place to another.

It essentially, eliminates a huge amount of advance planning and makes the whole trip smooth, easy and hassle-free. That’s what it’s like to use application-aware smart storage. Eliminate days if not weeks of planning, get the information you need when you need it, and get your job done by right clicking on VMs rather than wandering through a maze of LUNs, volumes, tiers and RAID sets. The benefits of smart storage are so compelling, that in a few years, we’ll wonder how we ever got along with it.

Edward Lee

Edward Lee is an architect at Tintri. Prior to Tintri, Ed was the Principal Systems Architect at Data Domain, and a key contributor to the first and subsequent releases of Data Domain’s file system. He was responsible for innovations like the BOOST deduplication protocol and replication. Prior to Data Domain, Ed was at Zambeel and Compaq Systems Research Center. He has a Ph.D. from UC Berkeley in Computer Science, where he was an original member of the Berkeley RAID team.