In November 2015, we released a Technical Whitepaper titled “Why Storage Architecture Matters.” It explored the need for high performance, flash-based storage systems and compared common architectural approaches found within the industry with the architecture of X-IO’s ISE and iglu solutions.
When I first saw the white paper, it was a bit overwhelming and a little too much to digest in one sitting. This series of blogs will breakdown the information in the white paper into smaller chunks to make it quicker and easier to follow. However, the full white paper is available here.
The first area we are going to look at is the often confusing and misrepresented question: “What do I need high-performance flash for and why?” By identifying use cases for flash, you can evaluate your environment for opportunities to meet demands for high performance.
Impact of Virtualization on Storage workloads
In a virtualized environment, storage systems support multiple applications that each have unique workloads. When combined, these workloads create a complex landscape of different block sizes, sequential versus random access, read vs. write mixes, etc. What this means for your storage system is that it is constantly dealing with random workloads and needs to be able to able to switch from the equivalent of unpacking a suitcase to unloading a semi-truck, all in a matter of milliseconds.
Database and online transaction processing (OTLP)
In a world where speed is key, businesses need to be able to quickly access their databases to provide both employees and customers with information. When slowdowns occur, companies waste money and customers grow frustrated; definitely not a recipe for success.
We already discussed how a storage system in a virtualized environment needs to be able to respond to random, complex workloads. Now, what happens when an application consumes more than its fair share of I/O? Well, think about that time you had a tight deadline and got frustrated with your computer’s response time and couldn’t get the job done. Super frustrating! There was likely another user or application taking way more than its fair share of computing power.
Virtual Desktop Infrastructure (VDI)
With VDI, the random, complex workload rears its ugly head again, this time, multiplied by the size of the environment. Let’s say you have 1000 VDI users and have built your storage system around an average of only about 100 users logged in at once. What happens when 300, 500 or even 1,000 users log in/out simultaneously? What about the other users on the system? How much are they going to impacted? Are the desktops refreshed when the user logs off, and how often are the desktops refreshed for patches? Does this have to be done during “off-hours”? What if you don’t have “off-hours?” These maintenance operations are one of many issues that directly impact VDI performance and demand a high-performance storage solution to help solve. Hybrid solutions, where all-flash performance is combined with affordable hard disk storage, help meet the various demands posed by VDI environments.
Cloud service providers need high performance because they are dealing with multiple applications using a storage system in a heavily virtualized environment, while providing services for hundreds of customers. Further, cloud providers provide service and availability guarantees to customers which drive up storage costs due to overprovisioning, etc. While performance is important, the amount of capacity that can be used (without causing performance problems) directly translates to the amount of revenue a system can generate.
After reviewing these areas, a few common trends stick out that demand high-performance flash:
Virtualization brings a bit of the unknown into your environment. Flash helps you be prepared for anything you might encounter.
- Complex, random workloads from different applications: Differences in block size and random writes/reads across the system require high-performance solutions.
- Low latency/response times: Some applications run great with <10ms of latency. Others however, require <1ms latency. What do your applications need and users expect?
Next time, we’ll take a look at modern storage architectures and compare them with X-IO’s ISE architecture. Stay tuned!