The Storage Hypervisor Part 2 — Flash Pools
In Part 1 of this series we talked about how ONTAP could be seen as a storage hypervisor and how the benefits of this could mirror the benefits of a compute hypervisor like VMware vSphere. The key ingredient is a common OS or platform from which we abstract from.
To give an example of this, NetApp is currently #2 in overall storage market share, but the ONTAP platform is the #1 storage operating system in use today, serving up exabytes of data (the number 5 storage OS is NetApp’s own Engenio line). And because of this common storage hypervisor, some pretty amazing possibilities come into play which can really position an organization for agility – let alone plain old value. And the benefits aren’t just limited to NetApp storage as the V-Series and the ONTAP Edge Storage Appliance can extend these benefits into more areas — but more on that in future posts.
In this series I’d like to first take a look at some of the unique capabilities inherent in the ONTAP 8.1.1 platform ranging from storage efficiencies, multi-protocol, scalable “infinite and immortal” volumes and more – and then build on this to show how these provide more value in everything from disaster recovery, private cloud, test-dev, and of course agility. Before I get into storage efficiencies, I thought I’d talk about a new feature just announced in ONTAP 8.1.1 – FlashPools. But to best understand FlashPools lets take a step back and look at various technologies as well as ONTAP’s unique way of processing writes.
“You Don’t Have [Technology-X] So Mine’s Better!”
So you frequently race cars with your friends and suddenly a new ACME turbo charger becomes available promising a 30% increase in horsepower. You run out and buy it, hook it up to your car and you’re feeling great about your new turbo-charged wheels. Your car is indeed much faster now. By extension it must be superior to anything else that doesn’t have the ACME turbo charger!
Imagine your surprise when you’re friend’s car which doesn’t have the shiny new ACME turbo charger is still keeping up with you. What happened? You thought your car was superior but now you’re not so sure. This mystery might only be answered by popping the hood on the other car and see how they do things.
Such is the issue with NetApp. There’s been criticism from certain quarters around the lines of “you don’t use flash for primary storage or automated storage tiering” without the context and understanding of how NetApp’s ONTAP handles I/O. At the end of the day what matters is performance and reliability.
Yesterday — WAFL (Write-Anywhere-File-Layout)
Most SANs have an NVRAM cache that the controllers write to. But ONTAP will journal the write requests to NVRAM. This method not only consumes less NVRAM and improves recoverability, but also improves response times, and allows for disk writes to be optimized. The NVRAM has two buffers — when the first is full, a consistency check will run to write all the entries to disk. Now unlike some other file systems, WAFL will store metadata within files, which allows much more flexibility into how to write to disk — hence “write anywhere” — significantly improving write performance. OK, this part was kind of technical but the take away here is that due to WAFL innovation, NetApp was always been exceptionally efficient at writes and didn’t always need fashionable technologies that others were implementing to keep pace. (for more detail on WAFL, here’s a whitepaper from 2006).
Flash Cache and now – Flash Pools
So FlashCache has been around for a bit now and is a part of NetApp’s Virtual Storage Tier. It’s basically putting a PCIe Flash card next to your controllers. The controller has fast access to the flash storage and can use it to effectively cache random read patterns. This can dramatically reduce latency and can for example increase the number of concurrent mailboxes that can be serviced by up to 67% in some tests. This provides excellent results for reads, and it also improves write performance by offloading some read traffic from the hard disk system. WAFL along with Flash Cache did a great job at delivering performance for years, but today the demands of the most write intensive transactional systems could benefit from something more.
New in ONTAP 8.1.1 is support for Flash Pools. A Flash Pool (using patented technology which is an extension to WAFL) is essentially augmenting a logical aggregate of hard disk drives (HDD) with flash SSD drives. But here’s the twist – the primary data stays on the HDD tier and never moves around (except for write cached blocks once they become “cold”). The intelligent algorithms will populate the SSDs with the most frequently used read data to accelerate reads.
At the same time, random writes will take place to SSD drives while sequential writes will use the HDD drives, allowing for the most effective use of HDD/SSD depending on the pattern. So from a write perspective the SSD drives are used to offload the I/O activity of random writes. Now some applications will intensively write, then read, and then overwrite the same data. FlashPool is uniquely equipped to service this type of data, as it will offer both read and write acceleration for this data. The read-write cache will automatically adjust for your workload patterns — set it and forget it. And Flash Cache and Flash Pool are designed to complement each other, giving you the option to experience the benefits of both working together.
Compare this to automated storage tiering – these systems will move your primary blocks around the different tiers, where as FlashPool is an extension to the WAFL technology which leverages the SSD pools for block caching and write staging. Some are inclined to think that a SAN without automated storage tiering system that’s moving your primary blocks around to different tiers is somehow deficient. Automated Storage Tiering can certainly be used to drive more performance, but so can Flash Pools. But before you write off a solution because it doesn’t have automated storage tiering, take a look under the hood and find out what’s really going on.
PERFORMANCE AND EFFICIENCY
At the end of the day the primary two benefits are performance and efficiency. Performance to drive faster reads and writes than were possible before. Efficiency to use fewer resources.
Consider the following scenario – a pair of FAS6210’s with 240 SAS 600GB drives for a total of 144TB. Now make the following changes – replace those 240 SAS 600GB drives with 216 SATA 1TB drives. Now add twelve 100GB SSDs into the flash pool. What’s the net effect?
According to NetApp this increased capacity by 47% and reduced cost by 23%. That’s a 50% reduction in cost on a per/TB basis and all while consuming 26% less power. What about performance? According to NetApp, IOPS did not change more than 2% (plus or minus) from the baseline, but that response times were significantly improved. In other words, improvements in both capacity and cost were realized, without any decrease in performance. That’s a nice combination to have when you can get it.
And of course what enables the Flash Pool? ONTAP does. Any SAN in the FAS family running ONTAP 8.1.1 or later can use Flash Pools. In the next post in this series we’ll take a look at some of the ways in which the ONTAP platform makes storage more efficient. Towards the end of this series I’ll shift focus into showing how the multitude of ONTAP (storage hypervisor) benefits can lead to savings and agility that parallels what VMware has enabled.