Firstly, apologies for the length of time since my last post. A combination of family, work and late vacation hasn’t left much time for blogging in the last few months.
The recent fun and games with Dell and HP nearly brought me back to the keyboard – okay, so the technology was wrong (3PAR, really?), but I did say HP needed to do SOMETHING about the EVA, so half points for trying, maybe?
I have to say though, I was at a Dell partner event recently, and the Dell guys seemed pretty relaxed about the whole thing – consensus seemed to be that they dodged a bullet (and a repeat of EMC’s black eye when they bought RSA for what the market thought was “too high” a price a few years ago) by not staying in to the bitter end. The market seems to agree with them (Dell shares have been rising steadily since they stepped aside), which I suppose in the end is the important thing really.
But while I was away, I had the chance to evaluate a number of up and coming technologies, one of the best of which was by a small company named Isilon Systems. These guys have an excellent story but seem to have been ignored by large sectors of the market (up until this week that is).
So I saw this product, went away with my mind full of the potential and started talking to my sales people, getting things lined up, and had half the draft of an article lined up which would have made me look like some kind of prophet if I’d published it last week (or maybe arrested on suspicion of espionage – but hey, them’s the breaks :-))
Because now, from being a complete unknown, Isilon are now in the spotlight as EMC’s next acquisition target, with a number of tech journals and media outlets twittering on about the “perfect fit” with EMC’s product set and how “complimentary” it would be.
So let me politely respond:
I don’t think so.
Isilon is in no way complementary to EMC’s current product line. It’s a competitor to many of EMC’s core products, and it has the capability to do to EMC’s lineup, what XIV has done inside IBM (think hot knife through butter).
To back up this assertion, let’s consider a few things:
- EMC’s mid-range technology is based on variations of the Clariion platform. This is the clunky dual-processor architecture from last century, being replaced in other forward-looking vendors by scalable grid architectures – spookily similar to the Isilon architecture.
- The Celerra NAS platform, the closest EMC currently has to Isilon, has spent 10 years developing from a “dire box of misery” (actual engineer quote) to its’ current, actually pretty usable form, but let’s be honest, it’s never going to move NetApp from the top spot. EMC need a revolution in this space, and the Celerra/Clariion combo is never going to be that revolution, no matter how integrated it gets. Isilon is built from the ground up for NetApp competitive takeout and forms a far more credible proposition.
- Data Domain, EMC’s other recent acquisition, uses a variation on the same Xyratex-manufactured storage server as its basic hardware. Rather than moving Data Domain onto the Clariion platform, it would surely be less painful to integrate the Data Domain architecture with the Isilon stack (this would also plug one of the more painful holes in the Isilon array itself – more on this later).
- Centera, EMC’s archiving product, is being left further an further behind, with software vendors providing better resilience and compression/deduplication and is looking old and tired compared to products such as HDS’ HCP product family. Replacing this with a product with the compliance capabilities of Centera and big disks would get EMC back in this game (similar to what NetApp have done with SnapLock – provide Tier 4 storage on an existing array, rather than force a new architecture).
If you strip away the Isilon File System, what is left looks suspiciously like a better version of my favourite product of the last year, the IBM XIV. Based on the same Xyratex hardware platform, Isilon has the same scaling of processor, cache and disk capacity that makes XIV such fun to design with, but at the same time, Isilon has some excellent unique points:
- 8Gb/s Infiniband back-end network
- Scale to 144 modules (nodes) in a single manageable environment – slightly more than XIV’s current 15 modules per manageable system limit – this is something XIV has promised since the original rollout presentation, but has yet to deliver.
- Virtualised equivalent to RAID 5 (N+1 protection level)
- Virtualised equivalent to RAID 6 (N+2 protection level)
- Virtualised equivalent to protection levels the market doesn’t have a name for yet – N+3 and N+4.
- Seriously, if there’s a snappy marketing term for these out there (other than “holy sh*t that’s a lot of protection”) then I want to know about it.
- Choice of disks and enclosure types – chose from SSD (SSD!!!!) SAS (SAS!!) and SATA (oh well) to give a choice of performance and capacity
- Diskless enclosure gives a performance boost without having to increase capacity – one of my big issues with XIV (#4 I think) was a need to scale capacity to get performance – if you wanted 50,000 IOPS, but only 5TB of capacity (actual customer of mine had this) then you bought 79TB and liked it (or else).
- Starting point of 10TB (3 modules) – technically you could start with less, but you don’t get parity protection till you have 3 modules, so this is the best place to start – still less than XIV’s 6 module 27TB starting point.
Of course, it’s not perfect yet:
- One of my more backup-focused colleagues poked a few holes in Isilon’s NDMP-based backup solution. Okay, not so much poked holes as picked up the argument-based equivalent of a cannon and started blowing it away. End result is – NDMP is unlikely to be fast enough to backup a moderate-sized Isilon. The fix? A conversion appliance that takes the IP traffic and converts it to FC (for backup only).
- This is where Data Domain comes in – combining the DD backup architecture with Isilon would remove backup issues.
- Isilon is currently file only – EMC would have to develop a block-level equivalent which can run in parallel with Isilon’s ONEFS file system, unless they wanted to do as NetApp and keep everything in the file space.
- To an extent this is feasible – with more enterprises moving to virtualised environments (courtesy of EMC again) the block vs file argument is becoming a little less important.
- Still, at least we know it would be possible to develop a grid-based block system to run on the Isilon hardware (thanks for the pointers, Mr Yanai!!)
- Isilon is not Fibre Channel capable for host traffic, but being based on Xyratex, FC capable HBAs can be fitted)
- Of course, by the time Isilon is integrated into EMC’s portfolio, FCoE may be the standard (he says hopefully) – again, the Xyratex servers can be retro-fitted with CNAs as required.
- Most important – please please please EMC, please hire someone to design a proper user interface? Maybe the guys who did the XIV interface could do something tasteful for you in blue?
A couple of years ago, Moshe Yanai told me that EMC would never develop something like XIV, as they were too wedded to the Symmetrix and Clariion architectures – too much development has gone into these platforms for EMC to contemplate spend money developing a competing architecture.
As this was based on Yanai’s belief that nothing like XIV existed to be acquired, I’d argue that a successful acquisition/integration of Isilon by EMC would blow that position away completely. Without spending time and money on the basic architecture, EMC will be in a position to develop a grid-based replacement, first for the Celerra then the Clariion, Centera and beyond (Isilon-MAX anyone? Just kidding, like the DS8000 range, Symmetrix has a number of features that Isilon can’t and probably will never copy)
In fact, if I were an EMC employee working in anything starting with a “C” I’d be looking for a quick change of career if this goes through.
Of course, as Dell proved, it doesn’t matter who makes the first bid, so much as who makes the last. I’ve listed potential counter-bidders in order of who I think is most likely to try and take this away from EMC:
- IBM – at the conclusion of the original presentation from Isilon, my first comment was “you guys are so getting bought by IBM” (seriously, this was my actual comment). I hadn’t actually considered EMC as a buyer (for reasons, see above) so IBM appeared the perfect choice – I don’t have a lot of faith in the SONAS stack yet, and Isilon would give IBM a properly developed scale-out file platform, with some of the features they’d like to have in iteration 3 of the XIV platform (Infiniband, parity protection, multi-rack scalability, choices). Plus it keeps EMC stuck with last-generation architectures.
- Dell – flush with cash from the 3PAR try, Dell could make a try for Isilon. Let’s face it, they’ve already hacked EMC off once this year by showing they want their own storage product, so why not go the whole hog just to see the look on Joe Tucci’s face? Some synergies with their existing storage lines – the Equalogic iSCSI array is based on a similar Xyratex storage server (seriously, is there anything out there that isn’t made by Xyratex these days?). At the very least they could do what they achieved with HP/3PAR and skyrocket the price, at least damaging EMC a little.
- NetApp – can’t really see the Synergies here, NetApp are unlikely to abandon WAFL, and could probably develop their own Xyratex-based or similar grid platform for an expanded WAFL file system in the future without hitting ideological barriers (of the kind that protect the Symmetrix architecture at EMC). All the same, they can’t be pleased at the thought of EMC acquiring a “NetApp killer” and may be looking for pay-back after the Data Domain debacle. As with Dell, NetApp may enter the fight just to push the price up and make things hot for EMC.
- Cisco – perfect chance to acquire a storage product to complement their UCS servers. Unlikely as they have a lot invested in the different storage collaborations, but possible.
or Oracle – who from the rumours, may wait until EMC acquire Isilon, and then acquire EMC. Not my favourite vision of the future, but depressingly possible.
I really hope EMC goes through with this (the Isilon acquisition that is, not the Oracle one). I’ve always enjoyed working with EMC (products and as an organisation), but it’s become harder to justify in the last year, when you know there are better products on the market – better service only takes you so far against better/cheaper/faster products such as XIV. Assuming EMC make a proper job of integration (and they’ve done reasonably well with some of their other acquisitions) this may be the thing to get them back on track.