Brian Jackson - Fotolia
When we look at flash and solid-state drives, it's understandable that we focus on their amazingly fast flash storage performance. All-flash arrays are delivering more than 1 million random IOPS, while a similar-sized box of hard drives reaches 10,000.
With an all-flash array, you extend the life and performance of a storage area network (SAN) without the need to add to or replace existing gear just to keep up with the server farm. Plus, the availability of IOPS opens new use cases, such as highly efficient virtual desktop infrastructure.
A common way to use these all-flash arrays is to replace just a portion of primary storage with flash and use it as a speedy tier for active data. Most users find that less than 50% of the data on their current HDD's primary tier is active, and that the rest is used for warm data.
With the smaller, but much faster, flash tier in place, there are many spare IOPS available. Flash arrays use these to compress and deduplicate data being downgraded to the slower secondary disk tier. The compressed data uses much less space, though the shrink factor is somewhat dependent on the use case and data type. Up to an 80% reduction in disk use is common, sharply reducing the cost of the secondary tier.
Flash storage performance and cost factors in the all-flash array
There's another factor at work here. Compressed data typically can be retrieved and expanded much faster than raw data, and uses less bandwidth to traverse the network. Together, fast retrieval and lowered bandwidth make compressed secondary storage much faster. This tiering and compression can lower overall storage costs compared with the all-HDD array configurations being replaced.
In addition to acquisition costs, the smaller configuration uses less power and space, and it is more reliable. All of these factors substantially reduce the additional cost of that all-flash array.
Now let's talk about the impact of those millions of IOPS on the data center in general. Most of us think in terms of clouds and virtual clusters, with containers adding a lot of debate about future directions. Using containers roughly triples the virtual instances in any server, meaning far fewer servers are needed for a given workload.
That would be a blessing to any cash-stressed business, but for the fact that those containers still require storage I/O. Three times the instances in a server with no change to IOPS doesn't cut it, and traditional HDD-based arrays are almost always maxed out. Bring on an all-flash array to increase networked flash storage performance, and having more than 100 times the IOPS means that, in fact, the equation has swung in favor of the virtual server farm. Even with three times the instance count, we have as many as 30 times the IOPS per instance.
Likely, the servers run much more efficiently when the I/O availability is increased. That means jobs get done in fewer instance minutes, leading to options to reduce the instance count considerably. Using fewer servers and containers to do the same work means real savings in hypervisor license fees, admin costs and networking. This comes on top of the money you aren't spending to buy additional servers.
In fact, the shrinkage in the needs of HDD storage, the reduction in server count for a given workload and the savings on networking gear, coupled with fewer software licenses and other operating expense reductions, could, in effect, pay for the all-flash array.
Here are a couple of ways to monetize this. One is to take on greater workloads without the need to buy more servers. A couple of years with much lower Capex would be the result. Alternatively, purging the data center of older, slower machines will trim operating expenses while maintaining current capabilities. Also, selling the older gear and using fewer licenses will offset the all-flash array costs.
How can all-flash arrays improve IT?
But there's even more involved here. Outside of IT, users worry about how long a webpage takes to load or the time required to query the ERP system. Being able to devote all of those millions of IOPS to server I/O, as opposed to background compression and similar functions, all-flash arrays will make users much happier with how IT is performing.
We shouldn't be surprised that all-flash arrays have become one of IT's success stories. Their ability to pay for themselves in so many ways makes them a compelling investment.
From a technical perspective, vendors have also done a great job. Most installations are as simple as joining the array to the SAN and adjusting some routing. Here the array caches the legacy HDD storage and gives it a boost. In "greenfield" installations, the arrays come with the software needed to set up tiering, so, again, setup is fairly simple.
All-flash arrays have mainly targeted the SAN, leaving filers and object storage aside. Clearly, this is a result of ease-of-use issues and the pressing need of the SAN-based IT shops to boost flash storage performance. This is starting to change, and we should see wider adoption in 2016.
Are there any caveats in using all-flash arrays in the SAN space? It's essential to map out the entire storage pool. Yesterday's primary tier with enterprise disk drives is now expensive bulk storage, for instance.
With compression and deduplication in most all-flash array units, this storage may no longer be necessary as slow secondary tiers get a new lease on life and a big capacity boost.
Another consideration is the migration of secondary storage to an object storage model, coupled with a smooth interface to public cloud backup and archiving. This will complicate tiering data using the all-flash array data services.
Adding all-flash arrays to the data center is not a snap decision, but the financial and functional benefits from doing so are difficult to ignore.
Making sense of all-flash array performance claims
A two-track approach to all-flash array
Better IOPS for your all-flash array
Evolution of flash-based storage arrays