It is pretty much the accepted wisdom in IT organisations that if you want something “faster, better, and cheaper” you’ll end up disappointed. It’s possible to achieve one, or sometimes two of these characteristics, but almost always at the expense of the third. It’s called the Iron Triangle, and program managers make their living balancing these demands and mitigating the trade-offs that almost inevitably result.
ITWire published an article recently from Paul Shaw, Hewlett Packard Enterprise’s (HPE) general manager of storage, about important considerations when moving to flash storage. Shaw has asked to come back to readers on the affordability and resiliency issues.
He asks the question (as HPE is a disk type agnostic vendor) “Myth or fact – can flash storage have performance, affordability, or enterprise-grade resiliency and data services?”
There’s a school of thought that says the same is true in the flash storage world – that you can have performance, affordability, or enterprise-grade resiliency and data services, but you can’t have all three. But is that true, or is it a myth perpetuated by storage vendors to cover gaps in their product offerings?
The primary appeal of flash storage is performance – solid-state arrays provide performance levels and order of magnitude faster than disk-based arrays. How much faster? A high-performance 15K RPM spinning disk drive can perform at best a few hundred IOPS at 3 - 4ms latency. A high-density flash drive can perform thousands of IOPS with sub-millisecond latency. A fully populated high-density array can perform more than 1M IOPS while maintaining that sub-millisecond latency.
So it’s pretty much a given that flash solutions will deliver performance, and many of them can do it at an acquisition cost comparable to high-performance 15K RPM spinning disk. In fact, the right configuration of high-density flash can be had for as little as $0.02/GB useable storage.
Flash provides additional cost benefits. In enterprise IT, it’s common to over-provision disk storage silos to boost performance. The speed of flash eliminates the need for over-provisioning. Flash performance also provides benefits on the server side, reducing wait times and allowing applications to run more efficiently. Applications can run on fewer CPU cores, reducing hardware, software licensing, and maintenance costs. Flash arrays can also reduce storage footprint by as much as 80%, with corresponding reductions in power and cooling costs.
A recent Wikibon study proved these benefits, achieving a 75% overall read/write performance improvement and up to 6x reduction in latency by upgrading an existing Oracle OLTP application with flash storage instead of Tier-1 disk storage. Wikibon found additional benefits including the ability to accommodate more users, enable developers ready access to full database copies, and provide significantly faster report generation. The total cost of ownership (TCO) of upgrading with flash was found to be a fraction of the TCO of upgrading with Tier-1 disk.
Wikibon found that as latencies dropped, server wait times also dropped, allowing for substantial core count reductions, resulting in a 60% reduction in Oracle licensing and maintenance costs and a 50% reduction in the three-year database delivery costs for the sample in their study.
Performance and cost factors have been recognised in the industry- we are now seeing flash go through the classic technology adoption curve as enterprises shift from 15K disk drives to flash for high-performance applications. Flash will continue to become more affordable, with at least one industry analyst predicting that by 2017, the effective cost of deploying flash in the data centre will be lower than spinning disk and that cost benefit will increase rapidly over time.
But there’s a catch. The affordability of flash is directly related to the use of in-line data reduction techniques like deduplication and compression. These techniques can easily provide compaction ratios of 4:1 to 6:1 and are the key to achieving the benefits of flash storage in the enterprise.
Unfortunately, data reduction is not a standard feature on many flash arrays – it’s often provided either through an external appliance, which can increase cost or through software, which can impact application performance.
And when it comes to the third side of the triangle – resiliency and data services – most vendors can only dream of providing the massively parallel architectures and Tier-1 data services like transparent failover and synchronous/asynchronous replication that you’ve come to take for granted with spinning disk.
A faster, better, cheaper solution
If you know what you’re looking for, you can find flash storage that meets all three criteria – faster, better, and cheaper. When shopping for flash storage, make sure you’re getting the whole package. Look for flash arrays that provide:
- Scalability to the petabyte level
- High-density drives (3.84TB SSDs) to minimise floor space and power requirements
- Performance up to 1M+ IOPS at sub-millisecond latency for a fully loaded array
- A price point comparable to or lower than 15K spinning disk (as low as $0.02/GB useable storage
- Hardware ASIC that offloads processing like deduplication and compression from servers/controllers giving these components more bandwidth to perform their primary functions
- A mature and massively parallelised storage architecture that performs well under failure conditions and provides proven six-9s availability
- Tier-1 data services that guarantee resiliency
- Data protection that allows you to combine the best of snapshot replication with off-array backup and archival
- Storage federation that allows you to pool multiple flash arrays together
All of this adds up to a flash storage solution that gives you all of the features you’re looking for without trade-offs. It’s a fact, not a myth. No more hard [disk] decisions about what to give up to get the performance of flash. Many vendors will try to sell you something less, but don’t be fooled. You can have flash storage that provides the best in performance, affordability, AND resiliency and data services.