Let’s take it from the top
What is behind the decision to fold Sun’s storage group into the server group? I’ll quote from Jonathan Schwartz’s blog and analyze.
Mr. Schwartz has hold of a serious idea here: if Sun were competitive in storage it would be a profitable and growing company and the stock price would be several times higher.
Sun has been the sick man of storage since I started watching them in 1993 (see my post of 3 years ago Sun’s sorry storage story). I worked in Sun’s storage group for a few years – ’95-’98 – and found the culture dysfunctional, an observation born out by the dismal results of the last decade.
Sun had a server attach rate – the percentage of Sun servers with Sun storage – of about 75% in 1993. Today the attach rate is, I’m told, in the ‘teens, meaning that a couple of billion dollars of revenue and a billion dollars in gross margin is missing from Sun’s income statement.
The text
[JS]: “I’m radically increasing Sun’s focus on storage today.”
[StorageMojo]: I’ve heard that before: Encore acquisition; MaxStrat acquisition; STK acquisition and a few more. Maybe it’s real this time.
[JS]: “First, I’m going to be combining our Storage and Server product teams to create a new converged group at Sun known simply as our “Systems” team.”
[StorageMojo]: Computer companies used to sell systems. Systems are the new black.
[JS]: “So we’ll still be strongly focused on being a multi-platform storage provider (just as our servers run multiple operating systems, and our operating system runs on every vendor’s servers), but we’re also going to start talking at a higher level to customers that see more standardization and integration in their future datacenters.”
[StorageMojo]: We think what Google and Amazon are doing, integrating cheap storage into Internet-scale systems, is the future. And we’re SOL in the present.
[JS]: “Now, why do I believe combining groups make sense? It’s a recipe that works for us. We combined our high volume x64 server group with our traditionally high scale SPARC server group over a year ago – leveraging the volume skills of the former with the scaling skills of the latter. What did that collaboration yield? ”
[StorageMojo]: Combining two successful engineering groups is one thing. Combining a failed group with a successful one is much more dangerous. Good luck.
[JS]: “Secondly, as our servers clearly show, we’re heading to a general purpose world – in which open and general purpose platforms will be the dominant drivers of growth, for us and the market broadly. ”
[StorageMojo]: Software is driving the market.
[JS]: “The first general purpose storage system from Sun was Thumper (our x4500) – powered by an open source operating system (Solaris), and file system (ZFS – soon to be parallelized by Lustre, a recent acquisition from Cluster File Systems).”
[StorageMojo]: Thumper is the first Sun engineered storage system in almost a decade that actually got people interested. And it didn’t come from the storage group. Hm-m-m?
[JS]: “Thumper rocketed to a $100,000,000 annual runrate within its first two full quarters of shipment (on a $13 billion dollar revenue base, that’s hard to see, but we certainly took notice – at least one competitor did, too). ”
[StorageMojo]: 10 years ago the A5000 FC array rocketed to a $750,000,000 run rate in two quarters – almost an order of magnitude better – when Sun was a smaller company. IMHO, if Thumper had been expertly marketed Sun would be trumpeting a $1,000,000,000 number. Sun still has some organizational work to do – to be fair, so does most of Silicon Valley – to get creative and passionate marketing for its products. (See Coolest New Box: Sun Fire X4500.)
[JS]: “I want to focus on one particular group, whose value only grows to Sun every day – our Tape and Archive business.”
[StorageMojo]: This is good news: StorageTek will remain a standalone group within the “Systems” group. (See Curse of the Mummy: Sun wrapped in tape.) STK wasn’t a good fit for Sun, but their expertise is valuable.
[JS]: “Combine these assets with some of our recent network innovations (like Magnum, the world’s largest Infiniband switch – which is not the smallest variant we’ll build, btw), the Crossbow community in Solaris – and it begins to look like we’ve got all the right ingredients to reinvent the datacenter.”
[StorageMojo]: The data center is sorely in need of re-invention, especially the storage piece. Who better to do it than the vendor with the least investment in the current paradigm?
The StorageMojo take
In most instances combining a server group with storage is a recipe for disaster: the server people resent storage margins and the storage people feel slighted. That could be the case here as well. But something had to change.
John Fowler, who heads the combined groups, should benchmark Sun storage performance against other system vendors. The numbers are dreadful – except margins – and a needed corrective for the curious air of unreality that has pervaded Sun storage for the last decade. Sun really has an important market opportunity: a seismic shift in data center architecture.
EMC is bearing down on this as well. Sun has no time to waste.
Comments welcome, of course. Comments are moderated, and since I can no longer access the spam file – no, I don’t know why – paste comments into the box rather than writing them there in case Akismet eats them.
“We think what Google and Amazon are doing, integrating cheap storage into Internet-scale systems, is the future. And we’re SOL in the present.”
Excellent observation. I recently spoke to a friend who works at Sun, and think I finally understand Jonathan Schwartz’s strategy. It is to target the “next big thing”, while managing the current thing and the installed base a well as possible.
“10 years ago the A5000 FC array rocketed to a $750,000,000 run rate in two quarters – almost an order of magnitude better.”
True, but the A5000 also caused most of Sun’s problems in storage today. The A5000 had serious reliability problems, in large part because it simply overpowered the capacity of early Fibre-Channel infrastructure. These problems disrupted Sun’s R&D away from the plan to create an Encore derived HW RAID front end for the A5000 JBODs. Even after the “Photon” matured with the A5200, as an uncached JBOD, it was unsuitable for OLTP ERP databases, significantly disappointing customers.
“The data center is sorely in need of re-invention, especially the storage piece. Who better to do it than the vendor with the least investment in the current paradigm?”
I 100% agree.
I think the future of data center networked storage is filesystem based, specifically parallel NFS. I think technologies like FCoE are more likely to be enabling technologies for pNFS than a primary access technology on their own.
I believe Jonathan Schwartz has a “crazy like a fox” strategy. I still think Sun has significant shortcomings in some areas, but I also think the ball is in Sun’s court to become a major player in the next evolution of storage.
Mark,
I like your analysis. I hope Jonathan gets the time to make the strategy work.
As for the A5000. . . .
I was the product and marketing manager for the A5000. I wrote the product requirements document – which included a requirement for a RAID controller – but the decision was made to use Veritas software RAID. At that point, before the OpenVision acquisition, the Veritas guys had very little knowledge of the enterprise and were pretty cavalier about testing. The RAID controller would have saved a lot of headaches. Sun did rush out an s-bus write cache card, but that wasn’t much help.
Later I and the engineering team tried to get a switch option for the A5000, essential for tracking down problems in a big FC-AL system. Again, no dice, only a little hub that carefully propagated any problems.
I have to differ with you on the Encore front end – maybe someone entertained that idea, but the Encore code never got beyond Alpha quality – as the return of every unit proved. I worked on a similar idea using Sun servers, but the I/O limitations of the servers made it un-economic, as I think IBM’s Shark guys finally concluded as well.
The storage group had multiple opportunities to build on their FC leadership and refused them all. Why a company with the hardware expertise required for SunFire and the software expertise for Solaris couldn’t build a decent storage controller has always been a mystery to me. The A5000 problems were a symptom of a deeper problem: the inability of upper management to make good business decisions. I lay it at Scott’s door – he just never got storage despite all the money it made Sun. Weird.
The good news is that they have a second bite at the apple and, other than EMC, are the best positioned to go after it in a big way.
Robin