Around the Storage Block Blog
Find out about all things data storage from Around the Storage Block at HP Communities.

Flash-optimized storage design requirements: SSD vs. HDD

By Priyadarshi (Pd), Sr. Product Manager, HP 3PAR Storage

 

Pd - Copy.jpgToday, I want to continue the discussion I started on the the basics of flash-optimized storage—and dig deeper into design recommendations for a flash-optimized architecture. To design a flash-optimized system, it’s important to first understand how flash-based solid-state drives (SSD) as a media are fundamentally different from spinning media or hard-disk drives (HDD). 

 

Here I’ve summarized the key differences between HDD and flash media in the table below and have aimed to net out the architectural implications emanating from them.

 

table.jpg

 

The “where” and “why” behind flash design requirements

The characteristics around IO density and latency of flash (as compared to HDDs) reveal that a flash-optimized system must be built on an architecture that has an optimized I/O path to get the best performance possible. We call it the performance acceleration requirement for a flash-optimized design.

 

Next, if you look at the table, two things stand out. Flash is still relatively expensive as compared to HDDs on a raw $/GB basis. The write cycles available on a flash device are very limited as compared to HDDs. From an architectural standpoint, this means that we must be very efficient in using flash as a media. We call it the efficiency optimization requirement for a flash-optimized design. And while performance and capacity efficiency optimizations are necessary, they are not sufficient to make a system ready for real-world deployment. After all, would you ever want to tradeoff reliability and availability for performance?

 

When your data becomes unavailable or gets corrupted, the promised million IOPS mean little, don’t they? This is why system resiliency is a sine qua non for a deployment-ready flash solution. And finally, while most, if not all, vendors out there are busy creating yet another silo in their customers’ datacenters, flash-based solutions will get greater acceptance from storage admins when they don’t have to manage an island of flash storage. To the end, data mobility allows seamless and transparent data and workload movement across storage infrastructure is the final frontier for a truly optimized flash design, built around solving your real challenges, not creating more problems for you to manage.

 

Are you considering flash storage?

Check out this infographic explaining the four key design requirements for an effective flash-optimized solution and and how all solutions are not created equal.

 

 

Discover2.jpgInterested in hearing more on flash, software-defined storage and other hot 2013 storage trends? Join us on June 11th at 8:30 AM Pacific LIVE from HP Discover in Las Vegas to learn how HP Converged Storage is transforming the industry.

Comments
nate | ‎05-29-2013 05:25 AM

You forgot to mention a couple things.

 

The impact of software on top of the flash. e.g. using compression and/or de-dupe to improve flash's cost structure(it's probably not the best idea where latency is absolutely positively critical, but in an area where you don't mind getting say 5-10x the IOPS at a latency that is at very least no worse than spinning rust it's a good trade off to have - actual latency numbers I don't know specifics but would be surprised if it was that high regardless).

 

Most of the newer flash startups incorporate one or both of these as their main strategy to improve the cost of using flash (obviously).

 

Not knowing their performance metrics, I was just looking at some load balancers as a comparison, since they too provide in line compression for HTTP for example. And an dual proc 8-core Intel processor can among other things provide roughly 15Gbps of compression on the Citrix Netscaler platform.  dual proc 6-core intel on F5 can do 12gbps compression (among other things..).  So that's over a gigabyte/second of throughput of compressed data(I recall my last 3PAR T400 with 300 drives rarely doing more than 500 megabytes/second on the back end disks - so being able to do a gigabyte/second on a smaller array I believe would be a big win), often time data compression ratios can be pretty decent so maybe your looking at upwards of 4-5GB/sec capable of flowing through such a system.

 

I have no doubt such a topic has cross the masterminds of HP storage since it is really the main driver of improving the cost structure of flash (outside of using flash purely as a cache). I believe even more so than newer types of flash that are trying to overtake SLC.

 

There has been gzip available as hardware expansion boards for some time (just found this one with a quick search - up to 10Gbps throughput). Though I think perhaps lzo(?) might be more common compression for storage, lighter weight maybe?

 

in a nut shell.. you should cover the topic of compression and dedupe! Though I think compression would give more bang for the buck(just ask your fellow friends at HP Vertica), with dedupe mainly helping with VDI.

Priyadarshi(anon) | ‎05-30-2013 04:32 AM

Hello Nate,

 

Compression and dedupe ("compaction" technologies) are elements that I put them into the bucket of storage "efficiency optimization". 

 

It is enlightening to see how you immediately zero in on the associated possible performance/latency penalty that can come with most of the current compaction approaches. It is a very valid point, one that we have heard from numerous customers. For this reason, our approach is to delivery capacity efficiency in a manner that does not come into the way of superior performance.

 

There are some fundamental changes we have made in the 3PAR architecture to optimize the way it works with flash - e.g. the way we do sub-cache page writes. I will leave them for a future blog post when I start talking about 3PAR architecture. And by the way, you are absolutely right in guessing (?) that we are indeed looking at de-dupe and compression - that's roadmap stuff for now. And while I am at it, let me add that we are actually looking at other ways to bring down the cost of flash for our customers. Oops, more roadmap. Thanks for sharing your thoughts and always eager to hear more from you.

Leave a Comment

We encourage you to share your comments on this post. Comments are moderated and will be reviewed
and posted as promptly as possible during regular business hours

To ensure your comment is published, be sure to follow the community guidelines.

Be sure to enter a unique name. You can't reuse a name that's already in use.
Be sure to enter a unique email address. You can't reuse an email address that's already in use.
Type the characters you see in the picture above.Type the words you hear.
Search
Showing results for 
Search instead for 
Do you mean 
About the Author
This profile is for team blog articles posted. See the Byline of the article to see who specifically wrote the article.


Follow Us
The opinions expressed above are the personal opinions of the authors, not of HP. By using this site, you accept the Terms of Use and Rules of Participation