This week Nimble Storage announced an all-flash array in a series of events around the globe that were based on presentations at the main event in San Francisco. Many bloggers were flown in specially from around the world, to help get the all-flash message out. Nimble has been a long-time provider of hybrid storage solutions and has traded publicly since December 2013, after IPO on the NYSE. Moving to an all-flash architecture is an interesting move for a company that even two years ago seemed to think that hybrid was really the only way to go (read this good interview article in Techworld). So why all-flash and why now?
The architecture of Nimble’s storage platform is based on a feature called CASL or Cache Accelerated Sequential Layout. The company doesn’t seem to talk about CASL much any more, however you can still find a deep dive video on the technology on YouTube. To summarise, CASL converts random write requests into sequential writes by coalescing the updates in NVRAM before committing to hard disk. Flash storage is used to cache active or hot data based on adaptive algorithms that make intelligent decisions about what needs to be moved immediately to cache for subsequent re-read. Anyone familiar with NetApp’s original Data ONTAP architecture will recognise this approach and in fact in March 2015 Nimble finally settled two outstanding lawsuits from NetApp over alleged misappropriation of confidential information and poaching of NetApp staff.
So CASL is designed to work with disk media by committing data in large sequential writes. Flash is only used to service reads because at the time of designing the platform, it was assumed that flash would wear out too quickly. The two potential issues for CASL performance occur when there are large numbers of writes or if data being read is not in the flash cache. In the first instance, performance can drop as the NVRAM flush process has to wait for data to be written to disk. Sizing of NVRAM therefore becomes critically important. Second, the ability of the read cache to be effective again is determined by size; if the read cache isn’t sufficient, then data has to be read from disk and at that point the I/O latency increases dramatically.
As a design, CASL is an elegant way of continuing to use hard drives and (what was at design time) expensive flash. However since the company was founded in 2008, flash technology has advanced significantly. At the macro level, prices have continued to drop (in the same manner as HDDs) while capacity and endurance have increased. At the micro level, new technologies like 3D-NAND and TLC have allowed the development of drives with capacities already at 4TB, with the prediction of 16TB drives within a couple of years. So it’s becoming increasingly practical to use lower endurance, cheaper flash as the bulk storage medium in todays storage appliances. It’s clear that Nimble have needed to follow the market and release an all-flash version of their Adaptive Flash platform, if not to compete on technical merit, but simply from a marketing position. All-flash systems are increasingly attractive, with many vendors pushing (effective) capacity prices down to around $1-$1.50/GB, so Nimble has to have a story, even if the original premise for the platform design predicated that all-flash wasn’t necessary. Incidentally, Nimble have not quoted raw or effective $/GB system prices, but instead chosen to talk about TCO (which is 33-66% lower than competing products, apparently). This in itself speaks volumes.
If we look at CASL, the use of flash as a cache becomes somewhat redundant with an all-flash system. Reads can simply come from the media itself. When processing writes, the NVRAM flush can be synchronised to match the size of the blocks of the underlying NAND. So, rather than have garbage collection continuously managing the block/page allocations, flushing to match block size should provide a more consistent performance experience and reduce the impact of wear on the flash. As the write process is well controlled, this should also mean that CASL is good for working with larger capacity drives (specifically greater than 4TB) without too much of a problem.
The Architect’s View®
There are some interesting aspects to Nimble’s technology, including InfoSight and some features of the all-flash systems like Triple+ RAID and rack density. These will have to get covered in another post. With some of those features, Nimble have moved ahead of the competition, however it won’t take long for the other all-flash vendors to start using 4TB drives in their systems. This week’s announcement looks more to me as a catch-up than a deliberate move to wait until the market was right for an all-flash Nimble system. However the markets did react quite favourably to the news with an 8.5% increase in the Nimble stock price to around $7.20. Remember though that the share price peaked at almost $53 in February 2014, so there’s a long way to go to get back to their former glory.
There are a large number of videos to watch on the Nimble page over at the Tech Field Day website. Alternatively, the links below provide other information.
- Nimble Storage gains more than 60 percent after IPO (San Jose Mercury News, retrieved 24 February 2016)
- Flash storage cost claims need ‘careful examination’ (TechWorld, retrieved 24 February 2016)
- Nimble Storage CASL Architecture Technical Deep Dive (YouTube, retrieved 24 February 2016)
- Nimble Storage Announces Settlement Agreement with NetApp (Nimble Storage Website, retrieved 24 February 2016)
- Nimble Storage Leapfrogs Competition, Unveils Predictive All-Flash Arrays (Nimble Storage Website, retrieved 24 February 2016)
- The Move to 3D-NAND for All-Flash Storage Vendors
Copyright (c) 2009-2022 – Brookend Ltd, first published on https://www.architecting.it/blog, do not reproduce without permission. Post #f748.