Starboard Storage Blog

Hybrid Storage SSD Caching Explained

By now, readers of this blog will know that Starboard Storage uses SSD as both a read and a write cache in order to offer the performance of solid state at the economics of spinning disk. One of the questions I get asked is “What magic do you use to promote data to the read cache?”, or “How does your SSD caching work?” To help answer this I like to use the analogy of a frequent flyer program and here’s why.

Let’s make the analogy that data is a passenger. There are four things we look for in understanding how to promote and maintain data in the read cache:
• What data needs to be accessed
• What the quality of the data is
• How recently the data was accessed,
• How frequently the data is accessed

This broadly equates to a frequent flyer program that tracks:
• If you are potentially a profitable customer
• How far you fly & how much you spend (quality)
• How recently you flew
• How frequently you fly

Essentially, data that is promoted to the read cache is tagged with information that enables us to track and assess its importance. Just like an airline keeps records of your flying habits we keep a record of your data’s habits. Unlike other SSD read cache implementations, we do not simply promote the most active or most recently accessed data and, as the cache fills, evacuate the first data that hit the cache to make room for more. We intelligently decide which data should stay in cache based on the importance of the data and flush the data that is least important.

Let’s go back to the frequent flyer program analogy. You may be a very frequent flyer travelling every week for a few months but that does not necessarily mean you are the best passenger for the airline. You might only be doing short hops for a job assignment and then never travel again. Another customer may be travelling once a month first class to Singapore and be spending more money and be a longer term “hot’ customer. The airline knows this and designs their program to enable a quicker path to the top tier of their program for the more profitable customers. It also extends the tracking period to a reasonable length, so the customer does not get bumped down if he does not travel for 6 months. Similarly if gets bumped down, he can quickly regain his status if he is a “quality” customer. He does not have to wait until the next year to be bumped up. He will be promoted the moment he hits the thresholds being tracked. The quality customer gets to the top tier faster, stays their longer and regains a lapsed status more quickly. We treat your important data in the same way in our caching methodology.

The concept of quality is extremely important in almost all analytic analysis. Big Data analytics is often used to correlate data to find specific patterns and traits that lead companies to their customers buying patterns and ultimately how to target better and become more profitable. When it comes to SSD read caching, Starboard Storage is treating your data like a “hot” customer analyzing which data to target for acceleration to make your overall system operate more effectively and make your company more productive.
In a future blog I will talk more about write-back cache optimization.

Footnote:- I recently read a blog by an all solid state vendor where they questioned the value of a hybrid storage methodology and tried to postulate that caching was a poor implementation for data and that customers should put all of their data on solid state. I will not call them out here because they did not let me post a reasoned response on their blog, but this is pure FUD, I will however extend my analogy to position all solid state solutions.

First let me say there is a place for an all-solid-state solution when you have a very specific and very demanding workload that you want maximum performance for. However, if you need 200,000, 300,000 or 1 Million IOPS, that is analogous to travelling first class on Concorde. It may have its place, but you are not going to fly all of your employees that way. You need to make sure that you have the appropriate business alignment for the value of the data and that is where a unified hybrid storage system such as Starboard Storage excels.
Of course, as you may know already from other blog posts or if you heard me speak at an event (check out http://www.starboardstorage.com/news_and_events/events/ for an updated list), with Starboard Storage, NAS flies free.

 Part two of this Blog explaining the Write Cache can be found here  http://blog.starboardstorage.com/blog/bid/277526/Hybrid-Storage-SSD-Caching-Explained-2-The-Write-Cache