Flash, flash, flash. If some vendors were to be believed, the obvious thing to do with any and all storage would be to throw out the tape, dump the direct attached storage (DAS), jettison the disk arrays and convert them all to flash. Such a strategy may work for certain environments. But it could be expensive and unnecessary in many cases.
So when and where should flash be deployed to get the most out of it?
1. Know Your Workloads
Organizations predominantly buy flash for performance reasons. Given its expense compared to tape or disk, therefore, over-provisioning can be extremely costly.https://o1.qnsr.com/log/p.gif?;n=203;c=204660765;s=10655;x=7936;f=201812281308090;u=j;z=TIMESTAMP;a=20400368;e=i
“The only way to avoid this is to properly characterize your production workloads (read/write mix, random/sequential mix, block/file distributions, directory/file structures, etc.) and then run workload simulations against potential all flash arrays (AFA) products,” said Len Rosenthal, Vice President of Marketing, Load DynamiX, a company providing flash performance validation via workload modeling and performance analytics software or appliances. “Users can easily identify the breaking points of their AFAs in pre-production via such simulations.”
2. Not for Highly Sequential Workloads
While testing is a smart idea, there are some general principles that tend to hold up – at least for now. Flash is vastly superior to disk, for example, in random access performance. So economics aside, it’s a no brainer there in terms of speed. But there are other areas such as sequential performance where it may be less advantageous to deploy flash.
“For highly sequential workloads (e.g. backup jobs or media streaming) flash may not provide the right price/performance advantage,” said Josh Goldstein, Vice President Product Management & Marketing, XtremIO.
3. Flash for Big Databases
Jim Handy, a consultant with Objective Analysis, tells users that perhaps the single best use for flash is in database management, both SQL and noSQL. If the dataset is absolutely huge, and response time is important then flash arrays are the way to go, he said.
“For smaller datasets and other types of applications, a flash array may not be the optimum choice,” said Handy.
4. Flash for Virtual
Vaughn Stewart, Chief Evangelist at Pure Storage, added that as well as databases, flash does very well in other structured data sets like virtual infrastructure and virtual desktops.
“From a performance perspective, flash excels in these areas,” said Stewart.
Handy noted, though, that some highly-virtualized systems may be best set up with DAS flash along with caching software.
5. Don’t Be a Dedupe Dupe
Many flash vendors throw deduplication into the mix in order to justify the ditching of a disk arrays. But Rosenthal warns users that deduplication and compression capabilities can vary considerably. Therefore, testing of these functions should be part of the evaluation process or else you may be duped by false dedupe claims.
“As these data reduction technologies have a huge impact on the affordability of flash, it is essential to understand the performance impact of such technologies under different workloads and data content patterns,” he said.
6. Use Your Own Metrics
It’s easy to get baffled by numbers. Vendors throw them at you like ticker tape at a World Series victory parade. That’s why Greg Schulz, an analyst with StorageIO Group, advised users to stick to their own metrics, those that matter most in their environment and to their applications.
“Numbers highlighting record-setting small 1/2 K (512 byte), 4K or 8K sequential reads might be impressive, but are those applicable if your environment does a lot of 64K (or larger) sequential reads, and 16K random writes,” said Schulz. “Just like hard drives, some are optimized for reading or writing performance while others for large capacity and the same applies to flash today, so keep that in mind what are you doing or need to do, what's your environment and application profile.”
7. Back Up the Flash
If flash performance isn’t what it should be, the culprit may not be the flash itself. It’s a bit like adding Formula One tires to a Toyota Yaris. Perhaps they will make the car go faster – provided you only drive and straight lines and never have to turn those massive tires which are way too big for the compact Yaris – but they are unlikely to make the car go fast enough to justify the expense. The moral of the story is to back up flash with a fast performing architectural foundation.
“Fast storage needs a fast server, fast application and fast interface so if you have bottlenecks, you need to address those as well,” said Schulz.
8. No Pain, No Gain
It’s easy to get caught up in the hype, but the smart way to get the most bang for your flash buck is to implement it where the pain is greatest. Those pain points will be very real to users, IT, and perhaps even to the C-level execs who approved the POs.
“Know where you pain point is and use some flash to buy you time or productivity instead of listening to buzzwords about IOPS, bandwidth, latency and queues,” said Schulz.
9. Level the Playing Field
Testing is fine in and of itself, but not if the results are skewed. This can and does happen.
“Before any performance testing, flash arrays must be pre-conditioned (writing to all flash memory cells) so that true steady state or normal operation can be assessed,” said Rosenthal. “Without pre-conditioning, performance figures will be significantly overstated.”
Look for more articles on flash over the course of this month including a buying guide or two, as well as more data storage tips and trends. Stay tuned!
Photo courtesy of Shutterstock.