Nimble flashes the all-flash array as ‘intense’ consolidation period approaches Hybrid array supplier Nimble is going to introduce an all-flash array, joining other hybrid vendors in riding two horses at once, against a background of coming intense supplier consolidation in the storage industry. Nimble, Tegile and Tintri are the three main new-style hybrid disk/flash array vendors taking business away …
Couple of things that make me think this is a lot harder for Nimble than just replacing spinning disk with flash:
1. Dedup. I don't see how how their AFA can be competitive without it.
2. CASL wasn't designed for all flash, so does this mean a complete OS rewrite?
Seems to me the basic argument they used against legacy vendors trying to do hybrid when they (Nimble) came along was, "You're trying to make legacy code do modern stuff it wasn't designed to do." Aren't they in the exact same position now that the move to all flash has happened so much faster than they thought it would?
It's called a Hybrid FLASH Array for a reason. if Flash was an afterthought add-on to CASL like it is with many other storage platforms they'd call it a Hybrid Disk Array at which point your AFA argument would be valid.
As far as dedupe goes, I guess we'll have to wait and see before jumping into any conclusions. After all, if anyone knows anything about Dedupe it would the Nimble founders who are ex-Data Domain.
As mentioned below by another poster, dedupe isn't exactly something you want for normal production workloads. It really only suits certain types (e.g. persistent VDI/RDSH) and is only a by-product of expensive hardware. Flash is coming down in price every quarter, we are at the point where we can probably have AFAs large enough and cheap enough where it isn't needed.
CASL was/is designed for Flash. If you have a look at the actual arch of the controllers and system and how it handles write commits/reads etc. then there is actually all that much they need to do to make it work fine as an AFA. Out of the younger storage vendors, I personally think Nimble are one of the stronger ones. The next couple of years will be interesting in the storage industry in any case.
Dedupe can be extremely valuable to some instances and not to others. To not offer it does seem like Nimble is missing out - AFA or hybrid.
We have hundreds of VMs that are the same, cloned over and over. The dedupe like crazy. We have datastores that are 90% deduped. Aside from the space savings, these bits get cached using less space, so the performance is greatly enhanced on a hybrid flash array.
When we patch all the Windows boxes that have been prestaged with the patches, the storage array barely blinks since its all cached in the read cache.
I think its crazy that some arrays don't offer NFS. I have colleagues that have a very nice VDI setup in a hospital - all non-persistent desktops - very sharp. They are using an AFA for storage, but its FC only. Since they have an incredible amount of vm creates and deletes every day, they needed to start a script that runs the FC unmap command to get their space back in a thin provisioned lun. If they had NFS this would be a non-issue. But in their case, post process dedupe wouldn't help them a whole lot since they are constantly provisioning VMs. Inline dedupe might help them quite a bit, but its not offered on this AFA.
They also had a former employee create the unmap command to run on all the luns from all the vm hosts, and that was a pretty sort point they had to figure out since he left - but its another matter.
For anyone saying no company cares about how much things cost, I would love to work where you do, sounds great.
Disclaimer: I work for a storage vendor that has dedupe.
For VDI, if you can do space efficient clones, effectively you are non-dupe and net the same benefits as dedupe, which can be substantial. It's where the biggest saving can be made by dedupe/non-dupe and my understanding is Nimble net the saving with space efficient clones.
Away from the VDI niche I have seen the stats from >1,000 dedupe enabled arrays and the typical saving is nothing like the dedupe vendors claim. 4% saving is at least as common as 40%!!! Having said that it does have it's place with certain datasets/workflows. I'd be surprised if Nimble haven't added it by the middle of next year.
I am disappointed that compression isn't as discussed as dedupe - for many workflows it is much more effective technology compared to dedupe.
Disclaimer: I work for Nimble Storage
CASL is a Log Structured File system. Log Structured file systems maintain a log within the file system itself. Data and metadata are written to the disk, in the form of a log.
Do not confuse Log Structured file systems like CASL, with logging file systems. Logging filesystems track changes to the file system in a separate circular log or transaction log which is separate from the on-disk data. Logging file systems have been written for and optimized for disk and while they do write in free space they also fill holes when contiguous free space become scarce.
The principle of a Log structured file system is that once blocks are written they can't be overwritten until they are completely erased. Does this sounds familiar in the SSD world? Thus new writes or overwrites are written in Free space. This is done primarily to improve upon the write efficiency of the file system and reduce Write Amplification. Older or invalid, data blocks are garbage collected and erased which means CASL is not a write in place file system, is not a hole filling file system which also means CASL does inherently system wide garbage collection already. This process should also sound familiar as we think about SSDs.
Also CASL already provides variable block inline compression and inline zero block dedupe which means fewer back-end writes again minimizing WAF and extending endurance while minimizing wear. Additionally, CASL already writes in large segment chunks per drive corresponding to an even multiple of SSD blocks. So we do pay very close attention how we write because although our cache is used for read-acceleration we do have to write blocks to it for reads...
By the way, all SSDs internally use some form of a Log Structured File system today.
https://lwn.net/Articles/353411/
Cheers
Nick Triantos
Yes, both these things are weaknesses of Nimble. Their lack of a File protocol also has hurt them.
The Dedup comes into play when you are talking about $/GB of SSD vs $/GB of NL-SAS. They could get away with just compression with the deep and cheaps providing data persistence. At a 2:1 ratio which is what they typically get with just compression, it's going to be harder to make that $/GB number work vs competitors in the all flash space when there's no cheap 8TB drives providing the capacity.
CASL wasn't "written for flash". It's a great block OE and works really well, but writes coming into the Nimble never touched flash, as far as I understand it, writes are dealt with in DRAM (with a BBU to make it NVRAM) and then laid down in big stripes onto disk. The SSD layer was read caching only. CASL might work well for flash also, because of the way they aggregate writes into single large stripes to be written, but it's heritage is spinning disk entirely.
File protocol is less important, but it's always hurt them vs NetApp and Tegile. As an AFA, probably won't matter as much. As a hybrid, yes definitely it has taken them out of deals.
Nimble makes a great product, but it's going to have some major challenges vs Pure and Xtremio among the other players in the AFA space.
Nimble takes down Pure and EMC everyday with Hybrid storage based on performance and features. Since the CASL performance in based on CPU speed and not spindles, the movement to a faster CPU, and AFA could make there product amazingly small and fast on only a few SSD disk. Then you might have the ability to replicate such an array to an existing model array for DR. Sounds pretty cool if they can pull if off.
Dedupe is a technology designed to give more space to people who can't afford the space they need. Who really wants Dedupe on a production array? It's just an overhead and a product of expensive disks. It's not a feature is a workaround.
As for CASL, I'm sure the developers saw Hybrid tech was only ever going to be a stop gap measure and they've probably been developing an AFA for a while now. You don't just release a product to market and then stop developing.
We all know spinning disk is cr*p, It's about time we stopped messing about with it and started realising that rust is rust is rust no matter how you dress it up.
Odd comment.....I haven't met a customer yet who is not very focussed on price....
& depending on how dedupe is used, it can reduce the costs of the storage by a massive percentage: nothing to do with customers who cannot afford what they need!
With flash in particular, reducing that cost (or increasing the capability of what that solution can address) at negligible performance impact is hugely powerful.
We should hunt around more for those customers of yours who care not a jot about their cost!
I do find it interesting that when a vendor can't do something everyone else is doing they downplay it as unnecessary or even portray it as a liability - right up to the point when they start offering the same feature.
Not necessarily picking on Nimble here, but marketing from tech companies in general.
With the speed of CPUs today inline compression is an absolute 40-50% extra space gimme - why the heck would you NOT want it in 95% of cases (which is a lot)?
Dedupe is a little more esoteric, and despite claims by some vendors it is suitable for everything and can't be switched off, I've seen evidence to the contrary, and dedupe rates can crash below that of inline compression in some cases.
Erasure coding is something else that's going to throw you 50-60% more space for absolutely nothing, so anybody not considering that move could look expensive on the £/GB stakes.
We are definitely moving to the commodity flash with NVRAM/enterprise flash tiering solution as the New Hybrid storage solution.
Disk will still keep ahead in £/GB for a long time - when you consider longevity in the TCO between TLC flash and nearline SAS HDD - but primary storage is heading inexorably towards all flash, and faster than you expect.
I thought Nimble came out with all flash over a year ago, though looks like that was just an all flash shelf. Their website implies you can pin a workload into all flash if you want, so coming out with an all flash version I don't see as doing much for them other than perhaps to get into gartner's MQ for "all flash" or something. I'd expect the cost for having a few extra nearline drives in the existing solution geared towards all flash is a rounding error.
Disclaimer: I work for a storage company that is not Nimble
There's a difference on storage arrays between front-end functionality - like LUN creation, volume management, remote mirroring, replication, application integration, snaps, consistency groups, etc. and back-end functionality like RAID, dedup, compression, rebuilds, scrubbing, defragmentation, tiering, etc.
No one should be buying an array for the back-end functionality because these functions are inherent to most, if not all enterprise arrays.
Dedup for example is not end state functionality: it exists only to reduce the overall cost per terabyte for storing your data. Compression and tiering serve the same purpose. In fact, there are many workloads that don't provide a good dedup rate and you'll get better (i.e. lower) cost per terabyte from them.
Not all dedups are the same either. Some only dedup with the LUN / volume. Others within a consistency group. Few dedup across the entire array dataset. Some don't allow the admin to set exceptions for workloads that would be performance impacted by the activity.
So the whole reason that Nimble are so late to the market (so I'm told) is that there was an agreement with the ill-fated Nimbus that Nimble couldn't release an AFA (due to name similarity.
Has this agreement expired or has Thomas cropped his head up from wherever he's hiding lately and settled?
This was always going to happen and they would have road-mapped for this from inception.
It's barking to suggest that a modern storage company would have recognised the value of flash - but only built their core proposition around it working with yesterday's technology.
Nimble will do hybrid better than most whilst spindles are viable and then they will deliver all flash at least as good as the competition in parallel.
The thing is regarding Nimble as a hybrid, their current platforms are more than fast enough for the majority of workloads. In my experience, companies are moving to AFA because of the space savings and cost compared to the incumbent array mfg. The speed of AFA is so much that only the largest and most demanding workloads are truly needing them. While there certainly are those enterprises that require the ability to do 200k+ those I think are the exception not the rule. It seems to me that the majority of Nimble's sales will continue to come from the hybrid arrays and the AFA is for the smattering of people that really need more than the 150k IOPs that the CS-700 provides. That and if they offer the ability to replicate from an AFA to a hybrid without the need for a costly AFA on both ends. That to me would be a huge benefit.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
