3Par Bid War – 3Par Shareholders say thank you to Gartner
Wow, the Dell and HP bid war for 3Par an 11 year old storage company is getting interesting. Dell opened up to the public that it was going to steal 3Par for ~$1B and or $18 a share and now just a few days later the price is up to $27 a share for a company that has great technology but has not fared so well financially in the rough seas of the storage industry.
So is 3Par’s IP really worth more than two billion dollars. Maybe but my guess is there is something more important at stake here. In very large organizations IT executives and financial executives leverage Gartner and other analyst very heavily as an elite level CYA. If you’re managing 2-3PB of data the value of that data is extremely high (sometimes in the billions) so the coolest technology, the best ROI, lowest TCO, whatever the toy in cereal box doesn’t matter if your not considered an enterprise offering in the Gartner Magic Quadrant. That’s where this get’s interesting. If you look at the model that Gartner analyst put together you will see very few storage vendors make it into this category and for good reason. In order to qualify you need to have a plethora of attributes in your organization beyond just financial scorecards and technology roadmaps. Basically for a large enterprise to consider buying from a company they need to know it’s not only rock solid but titanium forged granite solid. Beyond the ungodly amounts of money forked over to Gartner to get these credentials an IT vendor also needs to have the credibility and reliability to stand up to it. Gartner needs to hold up it’s recommendation to keep collecting on their analysis right?
Think of it as your investment broker or financial adviser, maybe even mutual fund. If the picks are not good you are presented with risks and probably not be doing business with them anymore. So you’re the kind of person who looks at risks as a way to gain? Well so do large enterprises. Just not with their IT investments. Risks is something to take with marketing strategies, product launches, or sales techniques. Adding risk to get additional value in the IT infrastructure is frowned upon and costs careers because there is too much at state with making a mistake. A lot of sales guys use downtime costs as ROI calculators well this is where that matters most because seconds of downtime are measured and millions are invested to keep those seconds down. Wrong decisions can lead to a lot more millions in costs in the long term and lots of resumes hitting the market.
So how does that matter with 3Par? Well there are 3 other quadrants to this magic formula for storage vendors and 3Par has been trying to break out their current standing into the Enterprise credentials. This is the only way they’ll be able to break into the very large very lucrative enterprise market. Sure they have plenty of customers with PBs of data managed on their system but those customers are for the most part internet companies willing to take the risk in order to get more bang for their storage buying buck. Not saying they’re not happy either, I’ve met plenty of very proud 3Par shops who’d probably do a better job of selling the technology than 3Par. The reason 3Par or Compellent, or XIOtech, or … hasn’t made it into that quadrant has nothing to do with their stellar technologies but rather the maturity of the organization, the financial background, and their potential futures. You don’t see any rumors of EMC getting bought, or EMC, or for that matter HP (all of which sit in the sweet spot on Gartner’s tool) and that’s why they’re there.
So, if Dell, who is not in that quadrant with storage were to snag 3Par it’d only be a matter of time before they get there on their own with higher margins than reselling EMC solutions. This poses a major threat to HP who sits pretty low in the Enterprise quadrant today. Not to mention the fact that HDS is evolving and HP’s enterprise XP system is OEM’d from HDS. Now the picture begins to clear up. Dell is trying to move up the chain and secure more business in large enterprises where storage is dominated by HP, IBM, and Dell and HP is making sure 1) that doesn’t happen and 2) they have a solid plan to stay there without having to rely on a potential shifting evolving HDS. Not to mention that even HP’ers are getting tired of the mid market EVA array and it’s lack of luster in that range of products. I personally see this is a great opportunity for HP to kill 2 birds with one stone and think they’ll pony up the most dough to make it happen. Not to mention how many HP customers would welcome a refresh of technology to talk about with their business partner.
Thoughts, rants, comments? Drop a line and let me know how you feel and what’s your take.
Flash! Gordon
In the enterprise there is a swelling trend that starting to build a tremendous momentum right now. Improving performance and looking at cost per IO compared to cost per GB. There is a realization that as environments grow and condense with virtualization that performance is a burden that traditional spinning disks cannot handle moving forward. Well they can but that’s how flash and SSD are making their case.
In order to achieve enough IO to hit requirements of lets say a large database slammed by a large online web presence one would traditionally look at building out a large scale array full of 15K RPM FC or SAS drives. How many shelves are required to hit a daily spike of 40-50K rIOPS? Well at 150-250 IOPS per spindle on a 15K drive you do the math, it’s a lot of rack space. That equates to a lot of power, a lot of maintenance bills, a lot of cooling, not to mention a lot of noise.
On the admin side there is already a case for SSD and Flash. Getting the bean counter to pay for it requires an extensive sales process looking at the TCO and ROI which really starts to make sense with the above considerations. As the technology is wider adopted we’ll see price points start to fall and it’s a domino effect from there.
Goodbye spinning disk? Yeah, anybody remember those DataDomain bumper stickers “tape is dead”. It’s not dead… yet. Neither will spinning disks for a long time. I recall a meeting with Violin Memory and a customer of my in the greater Seattle area where the Violin Account Executive dropped a great line that could and probably does hold true, “We can get to the same price as FC”. This was at a great lunch meeting and while the customer is currently using FalconStor and Violin together there were quite a few laughs at the table that day. Enterprise customers are very concerned right now with price per GB/TB and it’s very hard for them to make an ROI transition. From a vendor sales perspective it’s clear as day, the writing is on the wall. SSD hits a better pricepoint per IO then spinning media. True, but from the user perspective that means tons of work building a case study to prove that it does. Not to mention that you’re opening a can of worms. Hmmm. So your telling me that the HVAC strategy you deployed in our new DC is ineffective? Why are we buying so much disk anyway? etc… This is a sure transition but it won’t be overnight.
So what options do we have in the meantime. Well there are three approaches I see my customers going to or looking at:
- Small amounts of SSD for “Tier 0″ data
- Flash as Cache
- Automated tiering of data between SSD, FC, and SATA
In my opinion, cache is the best option on the market right now. Obviously for one because it’s what FalconStor does with it’s IO Accelerator product and I have witnessed that solution increase performance of existing storage 10x with only a fraction of FLASH. Yes, a 2% ratio of flash sitting behind FalconStor NSS storage virtualization heads with existing storage consisting of FC or SATA drives with a 10 fold performance gain. Not to mention a 10x decrease in seek time which is more important in some instances. NetApp is on board with this idea as well. Nimble Storage, a startup which we’ll talk about more in depth on a later post is doing this, EMC looks to be migrating this way. It makes the most sense right now. Tiering and auto locating data is a challenge and it’s really a “post process” event. Compellent loves to talk about their progressive data teiring or optimized data placement or whatever the hell they call it but the reality is that an admin has to go in and define what a hot block is then when thresholds which are pre-defined are met data will be moved from SATA to FC or SSD or in opposite directions. That’s great except it only applies to data sets that get hot for extended periods and then go cold. Analytics maybe? What about IO spikes that occur daily, or unplanned database rampups. Well you’re going to be about 3-7 days to late getting your data into SSD with a system like that. Not to mention you’re taxing the system with additional writes. Copying hot blocks into flash with an intelligent algorithm like FalconStor’s HotZone is key. They’ve been doing it for almost 7 years now. On the write side there’s no argument. Write cache sizes in the TB is awesome. I’m sure there will be more technologies and approaches coming to a sales meeting near you but for now flash as cache is king.
Flash or SSD? Tomato tomaato? Not quite. They are two very different technologies and should be looked at in different approaches:
On the flash side I think the prominent players are Fusion IO, Violin Memory, and now Micron electronics. Fusion IO drops a PCI card into the server side in an attempt to decrease reliance on DAS/SAN side for IO. Obviously cramming a ton of memory into a DB server has the same effect but there are some pretty cool things going on with Fusion. Some of the challenges I’ve heard are that they’re card mitigates too much to the CPU eating up cycles. Micron’s new P320 card which looks to eventually hit capacities in the TB+ range claims to not touch the CPU and presents itself as a SATA device to the operating system. Pretty smart idea. And with a 750K IOP performance spec that’s a hell of a SATA drive! I’m doing my damnedest to make sure I can integrate that technology in my products ASAP. Violin has another take on the presentation of Flash. They have an array crafted of VIMMs or flash on PCI cards. This allows them to build arrays approaching 10TB in capacity with performance hitting 250K IOPS. They are a relatively new company with a lot of their staff coming from FusionIO actually. Great technology and I have a a few deployments of FalconStor NSS leveraging their array as a caching mechanism with SafeCache and HotZone. Challenge is their current connectivity is PCI-E which limits flexibility. Sure if you want to buy FalconStor, and there are plenty of reasons to take a look, it’s a cool idea. Other than that the only way I can see is setting the system up as DAS to 2 servers per array or presenting a NAS share off a gateway…
SSD is something I think most of us are more familiar with. Dropping standard drive size modules into FC, SATA, or SAS connectivity to provide an easy migration path. This is great and while there are a plethora of vendors with solutions: Intel, Mircron, Samsung, Stec, etc… Most arrays are not built for this approach. Some of my customers have flat out stated they are looking at going back to internal DAS with SSD for top end database servers? What?!? I’ll save that argument for later as I need to go back to work but I don’t see that holding ground long. SSD is typically much more affordable than Flash and in general primary storage use I’d imagine seeing it take up the lion’s share of the market in the long term. As for performance I’ve seen a few installs of SSD crammed SAN appliances hit extremely high marks. Not anything near the Flash as Cache approach (at least in my customer’s installs) but very decent and very affordable. Again I have to point out Micron with their P300 drive which is hitting 20K IOPS!!! Yes, and these things are affordable. Let’s not forget that Micron based out in potato land was KING of memory for a very long time. Is this their move into fame again? I sure hope so because those guys know how to crank out quality product at affordable prices and with some of their newer technology hitting the market it should prove to be interesting.
Crap I have to go and actually work. Peace out!