Original URL: https://www.theregister.co.uk/2010/03/17/pillar_fork/
Ellison's storage Pillar sits at fork in the road
Finance, founding and two choices
Comment Larry Ellison founded Pillar Data on the premise it would become the next EMC or NetApp, and go through an IPO or be bought. On its way there it sees a new system that could replace 100 NetApp filers coming by 2012.
This perspective on Pillar was given by its CEO and CFO at a customer summit in Phoenix, Arizona on March 10. CFO Ned Hayes said: "We aspire to go public at some point in time and we look at the IPO market very seriously from quarter to quarter." He also said Pillar faced a fork in the road, with an IPO on one side, and being bought by a large incumbent on the other.
Who? Any one of Dell, EMC, HDS, HP, IBM, Netapp etc with Pillar's Axiom technology and products replacing the existing mainstream product set. A Pillar spokesperson said the firm wasn't actually talking to any incumbent about such an acquisition.
Currently the IPO route is depressed in terms of numbers of IPOs, valuations of companies and the returns for investors after IPOs. It's not a good time right now, but Larry is in this for the long term. Hayes was giving out a message that Ellison does not want to pawn Pillar off to the market at the first opportunity but wait until the market comes to him wanting to buy it at a valuation he agrees with.
Hayes said Larry Ellison has pumped several hundred million dollars into Pillar through Taco Ventures in the past decade. It has not been funded like a standard start-up with four to six sequential rounds of funding from groups of investing organisations en route to a relatively quick IPO flip so they can get their return. Larry wants a proper valuation for Pillar and the funding support is there if the recession proves to be a double-dip one. Hayes said: "We have Larry's capital and can wait."
There is only one investor so board-executive relationships are simple and direct.
Ellison, we heard, did not buy Sun for its storage, the majority of which is OEM'd from HDS and LSI. Larry hates OEM deals and, pretty soon after the acquisition finalised, the HDS USP resale deal was canned. How long has LSI got? Oracle wants to build a great system to run Oracle and its clients, and the 7000 line is the one Oracle seems to be focusing on. Pillar wants to build a great storage system to run applications in general.
We can surmise that several hundred million dollars is north of $200m but not as much as $500m, otherwise 'half a billion dollars' would have been mentioned. Let's settle on $300m as a round number. The target is to get it to become self-sustaining. Cash flow break-even will occur this year or next. A customer asked Hayes how close Pillar was to its target gross margin (GM) - the reply was that "we're on the right trajectory. First sales generally have a lower GM. Later service, upgrade sales and repeat business have a better GM generally. Next-generation platforms are moving away from proprietary hardware to off-the-shelf components. That gets us much closer to our target GM." Pillar recorded its highest GM in 2009.
My thinking here is that Pillar needs the new development project kit to reach its target GM and that's not going to happen until 2012 or so.
The total addressable market for Pillar is said to be $30bn now, composed of hardware (HW), software (SW) and services, and $40bn in a few years time with a near-15 per cent compound annual growth rate, outpacing the storage market. Top line growth in 2010 is set to be 50 per cent upon 2009. Repeat business is said to be 60 per cent a quarter.
The company cut costs in 2008 when the recession struck and went down to a 325-person headcount. It's now up to 340 people and wants to hire more. Hayes said it is developing a new platform with radical, massive improvements in its storage characteristics.
Paso and Napa
CEO Mike Workman provided some more details but not to anywhere near the extent of pre-announcing new products. Instead there are two main development efforts which could bear fruit in 18 months time. One is called Paso, a re-invention of the Axiom 300 using industry-standard components instead of the proprietary HW. As a recap an Axiom storage array provides unified block and file storage with data tiering and quality of service features, and has a design with dual RAID controller storage enclosures or Bricks linked by Fibre Channel to I/O processors called Slammers. Bricks and Slammers can scale separately and are managed by a separate Pilot box. It's a step on from a dual controller modular array, like EMC's CLARiiON or HP's EVA.
Workman said the 300 was relatively expensive for the medium business market compared to products like Dell/EqualLogic or HP/LeftHand, as it had more enterprise features: "We think we need a lower cost base for that machine going forward." So it could be used, for example in branch offices.
Paso is seen as having 10 times the Axiom 300's compute power and 10 times the 300's IOPS rating, with 100,000 SPC-1 IOPS mentioned, up to 144 spindles and be half the $/GB cost and half the $/IOPS cost of the 300. It will also scale more: "[Extending] to what today would be eight Bricks."
The other development focus is a high-end machine called Napa. Wine-buffs will know value proposition wines come from California's Paso Robles area with the high-quality more expensive wines grown in the Napa Valley. Napa is "the one that will scale to huge numbers and replace 100 NetApp filers in terms of I/O and manageability, and still have just one SW license".
Napa is envisaged as having three times the Axiom 600's compute power, around 1,600 spindles, twice the Axiom 600's maximum, and hold 3PB of data using today's (2TB SATA) drives. It will have half the Axiom 600's production cost, and that will help Pillar's GM. When fitted with solid state drives (SSDs) Napa will output 1 million SPC-1 IOPS and Workman contrasted this with 3PAR's 200,000 SPC-1 IOPS from 2,000 spindles. It should have 400GB/sec of memory bandwidth between the controllers, will output two million RAID IOPS, and have 12TB of cache, 11TB of that being SSD with DRAM making up the rest.
Workman said: "There's nothing better in the world than instantaneous data progression." With this much cache, hot data can be stored in it from day one instead of having to wait for statistical programs to discover it's hot and then move it into the fastest tier of storage. Speaking of tiered storage in general he was quite caustic about NetApp CEO Tom Georgens' view that tiering was dying: "He is just plain wrong."
Then he said Napa is made up of Pasos, up to ten of them - "It's all the same stuff," - with "the controller in Paso becoming the head of string in Napa". These Pasos should be linked in a switched InfiniBand (IB) fabric. The IB speed wasn't mentioned. We have 20Gbit/s and 40Gbit/s IB products now and we could expect, hope or guess that 40Gbit/s might be needed for the performance Pillar has in mind, but 20 is faster than 8Gbit/s Fibre Channel and cheaper than 40.
More controllers may be needed
This is where we can get busy surmising again because no more architectural details were revealed. It seems to me that Paso controllers are busy running Pasos and that there might need to be a second level of controller, controlling the Pasos and presenting them as a single storage system. So we would have these Slammers, presumably reinvented ones, with the 12TB of cache, as the main controllers, linking across the IB fabric to the individual Paso string controllers which talk to their local Bricks. Perhaps these main controllers are in fact Brick-less Paso controllers? We'll have to wait and see.
The vision is that Paso can be upgraded to Napa with the addition of an IB fabric being a necessary part of that.
Workman said that none of today's storage controllers had been designed with SSDs in mind. Because of this they are poor at handling SSDs, limiting the number of realisable IOPS from them. Both Napa and Paso are being designed with SSDs in mind. EMC and every storage vendor have to redesign their storage controllers to take full advantage of SSDs. NetApp is facing problems and that's why it stopped after adding PAM accelerator cards to its controllers. Adding SSDs in the storage enclosures is causing it heavy problems in his view.
He said that a Napa with SSDs should cost 80 per cent less in $/GB terms than an Axiom 600 with SSDs.
Napa will extend the Axiom range at the high-end and Paso will do the same at the low-end. A whole lot of high-end features should be coming because of the new HW, with replication engines being mentioned.
The level of detail here was quite surprising - no one had to sign non-disclosure agreements and Workman livened up his presentation with salty, pithy comments about other suppliers and technologies. If you read his blog you get the flavour. Napa and Paso are not products. They are development directions. It's my understanding that the existing products have developments planned for them too.
Pillar is serving notice on its various constituencies of customers, channel partners and its competitors that it is no Copan or StorSpeed. Larry is not impatient for a return on his investment. He wants to build an 800lb storage gorilla that can stand on its IPO feet or else get bought by a vendor who uses its technology to replace entire swathes of its storage product lines.
Workman breeds wiener (dachshund) dogs but he has more of a pit bull type product technology in mind. Wiener arrays won't cut it, not with today's data growth and SSD technologies. To cope with those you need the 800lb gorilla canine and Workman hopes that Paso and Napa are two dogs that will worry the heck out of the storage supplier sheep out there. ®