Introduction
Storage is at the heart of virtualization performance. Often, if there is an issue with virtual machine performance, it comes down to improperly configured or over-provisioned storage. Storage performance is key to making sure a virtualization deployment on any hypervisor is successful. There are definitely important storage metrics that must be considered both in sizing for a greenfield installation as well as monitored in day 2 operations. These important storage metrics ensure a healthy, performant storage environment that is able to satisfy business-critical demands acceptably and leave room for growth.
- What are these important metrics for storage performance and how do each weigh into the performance of storage backing a virtualized environment?
- Are these metrics in and of themselves the important factors to consider?
Let’s take a look at important storage performance metrics.
Table of Contents
Important Storage Performance Metrics
We will look at the following important storage performance metrics and see how each are important when building out a storage system or a virtualization environment. Understanding how each affects the performance of the storage system helps to give us a clear picture of how the virtual machines running on top are affected.
- IOPS
- RAID level
- Latency
- Throughput
IOPS
There is perhaps not a more popular or well-known metric when it comes to storage performance than IOPS.
What exactly is the term “IOPS”?
The term IOPS refers to input/output operations per second. Often with the marketing that surrounds most storage arrays, this is one of the most touted statistics of any storage system or storage array model. Especially when dealing with traditional storage arrays utilizing HDDs only, IOPS were critically important as they were certainly a good indicator of the performance potential of the array in general. While we can’t discount IOPS ratings, today’s flash-based arrays can deliver IOPS ranges in the hundreds of thousands of IOPS per second range which can at least on paper go well beyond most customer’s requirements unlike traditional HDDs-based storage arrays.
IOPS are also subject to the type of test that runs to determine that number. Vendors like to throw the highest numbers of IOPs out there for marketing purposes, but without listing the read vs write %, latency, and other factors, IOPS can be a bit deceiving. The problem is, vendors can spin up IOPS claims based on their own benchmarks and spin those numbers as they want. There is a good resource for non-biased testing of storage systems by an agreed upon set of real world testing to be found at the Storage Performance Council. The SPC posts benchmark reports of various systems from different vendors using its non-biased testing techniques.
RAID Level
It is important to understand how RAID levels affect performance when used on a storage system. Most storage arrays allow the configuration of various RAID levels. Each RAID level generally has strengths and weaknesses when it comes to either protecting data or performance. One of the critical aspects of RAID levels when thinking about performance is the write penalty that each RAID level incurs. Each level of RAID has varying degrees of write penalties. Consider the following comparison of RAID levels and the various write penalty that each incurs. The write penalty is expressed in IOPs. In other words, in the example below for RAID 5 it takes (4) IOPS to produce a write operation.
| RAID Level | Write Penalty |
|---|---|
| 0 | 1 |
| 1 | 2 |
| 5 | 4 |
| 6 | 6 |
| 10 | 2 |
While absolute performance is not the only deciding factor as we would never run production data on top of RAID 0, it helps to understand the differences in performance when looking at various RAID levels and how each affects the overall write performance, directly related to IOPs.
Latency
Latency is key storage performance metric that has a major effect on virtual machine performance. Put simply, latency is the time it takes for a single I/O operation to take place. If a storage system can perform 1 million IOPS but can only product that 1 million IOPS with 100 milliseconds of latency, it will feel and perform sluggishly! The lower the latency, you wait less time for the I/O operation to take place and that equates to faster performance.
Especially, with latency sensitive applications such as databases, it is crucial to have low latency I/O operations. Databases do not bode well under high latency conditions. When designing storage systems driving virtual machines that run database applications, IOPS of course are important, but as mentioned, latency in today’s flash arrays is perhaps the more important measure of just “how fast” a storage array can perform.
Throughput
Another key factor in determining storage performance is throughput.
What is throughput?
When thinking about throughput, it is the measure of how many I/O operations are able to be carried out in a given time, generally in seconds. Many storage arrays are measured in MB/sec. There are also normally two different values for throughput – sustained and peak. Sustained as you would imagine is the measure of the amount of data the device is capable of constantly pushing through it. Peak throughput is the amount of data at times of “burst” that it can achieve. Both metrics can be important. Sustained throughput is generally the more interesting of the two when thinking about a storage system backing a virtualization environment. Generally speaking, you won’t be in situations very often where the peak throughput values will be put into use. However, the day to day operations will certainly benefit from the sustained performance a storage system is able to deliver.
Thoughts
When putting together the metrics described above, the more I/O operations a storage system can deliver at low latency, the better the storage system will perform. If anything, when looking at claims made by storage vendors, be careful to be wooed by awesome IOPS claims as these can be misleading.
When talking with sales engineers who toss out IOPS, be sure to ask what specific tests were used to achieve those numbers.
- What were the read/write ratios?
- What were the data sizes?
All of those questions can help to cut through marketing materials and give a much better idea of what real-world performance will be experienced with a particular storage system. Additionally, utilizing resources such as the Storage Performance Council can help to get a non-biased third-party evaluation of a particular storage system.
Follow our Twitter and Facebook feeds for new releases, updates, insightful posts and more.
