Moon Macrosystems - How to build a better Sun
Go ahead. Take a crack at it
Analysis You're a Sun Microsystems systems engineer. Or a Sun software developer. Or maybe even one of the few sales people that knows a disk drive from a processor from a TCP/IP stack.
You have survived countless layoffs since 2000, when the dot-com gravy train ran under the sink. You've rolled with the seemingly endless strings of quarterly losses or skinny profits, your stock holdings (i.e. retirement funds) battered. The promises of a better future have stayed there - in the future.
You have (sometimes) enjoyed the grumpiness of Scott McNealy and winced at the missteps and cockiness of Ed Zander. You have braved on through the tumultuous years and smug management style of Jonathan Schwartz, as the family jewels were exposed through open source, and private equity investors with relatively minor stakes determined the future of the company for their own capitalistic purposes. (Shoulda gone private last year when you had the chance and the cash, Sun).
And now Larry Ellison is getting ready to come in and very likely gut the company. So you're probably working on your resume, while at the same time running through your accomplishments in preparation for justifying your existence at Oracle to the new Sun King.
But what are you thinking, deep down? Not about how to keep your job - that's for damned sure. What you're feeling goes well beyond the fear of unemployment. No, my friend, I know exactly what you are thinking, and it is this: How am I going to get those bastards?
Trust me. You can get those bastards, and I am here to show you how. And it is not a very original play - quite obvious, really - and in fact, it is the same kind of despicable tactic that the Sun King himself has used. There's nothing better or sweeter in business than attacking someone with their own strategy, and that means it's time for a little something I'm calling Moon Macrosystems.
Let's start with the obvious. The Solaris server installed base might have contracted by a factor of two since 2001 - down to maybe some 1 million installed servers globally - but it's a good business and one worth chasing. And it's one that can grow, but not with the overhead that Sun has been carrying.
Sun was charging too much in the wake of the dot-com boom for iron, software, and support. And compared to its rivals in the x64 server racket - who push mostly Windows and sometimes Linux - Sun has been charging too much as well. And in the areas where it has been price competitive, it's not able to generate profits.
This is obviously not a sustainable business model. Hence Larry Ellison will make a very loud $5.6bn burp perhaps sometime in June or July, and the Sun that emerges from, er, the acquisition will be substantially reconfigured to help the Sun King buy a new yacht or whatever it is gazillionaires do with their dough.
Ask yourself this: What is Oracle really buying for $5.6bn? An enterprise customer list and some partners, a bunch of open source projects it could participate in and benefit from for free, some open source respectability, and a pretty big patent portfolio that might or might not be useful. When you put it that way, the Oracle deal sounds kinda stupid. But Oracle can - and will - extract money from those enterprise customers. But perhaps not all of it. There's some for you, if you act fast.
The software is the easy part
I have said time and again that Solaris is the secret sauce that has made Sun what it is, and that is as true today as it was a decade ago - or even two decades ago when it was called SunOS. Lucky for you, Mooners, Jonathan Schwartz has cut your new company a break by taking all of that Solaris software stack open source. (One could argue that open sourcing that software made Sun less valuable even as it made Sun more relevant in a Linux-friendly data center environment. It also made some customers feel a little more comfortable, since the Sun software stack can - and will - outlive the company, no matter what happens to it).
For phase one of Moon Macrosystems, all you gotta do is download the code from Sun's open source projects, start up a Moon.org project site, rip out the Sun or other project logos and replace them with a set of Moon Macrosystems logos for the newly rechristened software products. Now you have a software stack. It might look a little something like this:
- Lunaris - operating system (formerly known as Solaris).
- Loonaris - open source development version of Lunaris (and you're crazy if you put it into production)
- Dark Side - integrated file system (ZFS, and so named because file systems should be invisible)
- Armstrong - Trusted Lunaris Security Extensions, of course
- Buzz - application development language (Java)
- Craters - server virtualization (VirtualBox or Xen or Containers or LDoms - pick one. Really, pick one. And if you can't bring yourself to pick just one, pick no more three or four. Okay, five if you really must have vSphere. If you say Hyper-V, you're fired).
- Maria - integrated relational database (could be MySQL, but let's get a better storage engine than InnoDB that can handle all the threads in modern iron, please. Alright. Forget MySQL. Maybe just go to EnterpriseDB and use its version of PostgreSQL and use its Oracle compatibility layer to kick Oracle in the ass).
- Regolith - application server (GlassFish)
- Lunar Eclipse - forget NetBeans and just use Eclipse like everyone else for an application development framework. Put the Sun Studio compilers in here and get the Mono C# runtime environment in there as well as a PHP engine.
- Houston - system monitoring and management tools (a mix of DTrace and xVM Ops Center and the other systems management tools Sun has cooked up over the years)
- Full Moon - high availability clustering. No branding changes needed here (funny, that)
- Werewolf - MPI parallel supercomputer clustering
- Lustre - parallel clustered file system for supercomputing, running on Lunaris
- Blue Moon - integrated hierarchical storage management (doesn't existed in Solaris (as far as I know) when using ZFS. But SamFS certainly was an HSM, and this kind of functionality is needed to stage data from tape and disk to flash and main memory automagically)
That is by no means an exhaustive list, but you get the idea. That was the simple part with the software stack. Now comes the hard part. The Lunaris operating system and all of these parts have to be tightly integrated and supported as a single stack. That means a set of user interface screens with a consistent look and feel so you don't drive system administrators mad. It also means that the entire stack has to be fitted to take advantage of all of the features in Lunaris. So the xVM Xen hypervisor has to be DTraceable, like the rest of the stack, for instance.
Learn your Big Blue lessons
You could learn some lessons from Big Blue and its System/38 minicomputers from the late 1970s - and their kickers, the AS/400s from the late 1980s. Cisco Systems certainly is learning this with its "California" Unified Computing System, but Cisco still doesn't quite get it, just like DEC didn't with the VAX and Hewlett-Packard (sort of) got with the HP 3000 minis.
But you also have to avoid repeating some of the mistakes these machines embodied, mistakes that lead to the decline of proprietary minis and the Unix boxes that followed them to market. They've largely been replaced by Windows and sometimes Linux for new applications. (Legacy Unix apps are doing relatively OK out there because companies are loathe to change platforms when something is working. Changing platforms to save a little money can cost a lot of money and a huge amount of hassle. So there's no point).
The first lesson is that integration means a system. Moon Macrosystems is selling systems, not servers, not storage arrays. Systems, when constructed properly, can deliver a lower total cost of operation. Ease-of-use matters, but you can't rip the tits off the cash cow, like IBM has done with the AS/400 by overcharging for integration and overplaying its hand on legacy lock-in.
Moreover, integrating software components does not mean selling the whole stack as one giant hairball of code. The Lunaris stack has to be licensed so any component can be turned on or off at will - on a physical or virtual machine - and customers must only pay for what they use. Find a pricing model that makes sense. You might price software by memory usage per month and leaving it at that, getting around the whole physical-virtual divide and using that pricing to differentiate the stack in the market. Get that annuity revenue stream that IBM is bragging about all the time.
Integration also means lower cost of support, as Ellison pointed out during the Sun acquisition announcement. But that lower cost of support only happens if you integrate and simplify the stack. Moon Macrosystems will support its own stack, which can run any Java, C++, C#, PHP, or Fortran application. That is enough of an addressable market. Honest. If customers want to run other databases or middleware or application development tools on the box, you will charge extra for this support.
The trick is to simplify the support enough on the integrated stack that you can charge a lot less dough for it. And while you're at it, get the good people at Sun who know how to do this support. This is not a deal you can outsource to Asia. You need the best of the best. Leave Larry the rest. You need automated support, something like the Lunaris Network, and an application exchange, like the Lunaris Exchange. You can start by merging Blastwave, the open source community for apps compiled for Solaris, into Loonaris.org.
The other lesson to learn from the venerable AS/400 - one that DEC and HP managers didn't get about their own minis and a lesson that Big Blue, to its chagrin, eventually forgot - is that an integrated system is not about itself. It's about the applications that run atop it. You make the system invisible, easy, beloved, reliable, and you get as many real-world, enterprise-class applications as possible on the machine and you help software houses make money by helping them push complete, turnkey systems to customers who are sick of thinking about computers.
At its peak in 1998, the AS/400 had 275,000 largely happy customers worldwide, served by some 8,000 software houses that had over 20,000 applications and drove some $5bn in server and storage sales for Big Blue. A decade later, the AS/400 is under $1bn in sales, and there are maybe 2,500 software developers on the platform and maybe 5,000 applications.
The hardware is the hard part
Of course, the underlying systems have to be called the Application Software System/500, tipping their hat to the integrated minicomputer from Big Blue that was the best selling back-end system for SMBs in the world for like 15 years and that's now the inspiration for the kinds of integrated systems that many server makers are espousing but not actually building - not yet, anyway.
You can take a number of different approaches to create the ASS/500 product line. One is to go to a company like Super Micro and just take the best pre-made, whitebox systems it has to offer and put the Moon Macrosystems brand on them and be done with it. Or, you can do what Unisys does, which is to resell Dell and Sun machines (after scratching the Dell and Sun labels off) and focus on its own high-end systems (which in the x64 space are actually co-designed with NEC).
Or you could get even more clever and do a deal with Avnet or Arrow Electronics, the two master resellers of servers in the world, and work to certify the Lunaris stack on a collection of key servers made by IBM, HP, and Dell and co-brand these machines with the Moon Macrosystems and Lunaris logos and get the hell out of the low-margin, high-grief hardware business entirely. And if customers want Sparc-based machines, do what Sun did: Cut a deal with Fujitsu. Trust me, the company needs money and if you can sell a box, Fujitsu will give it to you for a cut of the action.
One thing is for sure: Moon Macrosystems can't afford to do Sparc processor and system development, but you might be able to buy motherboards from Oracle or Fujitsu. And with IBM buying Transitive's QuickTransit emulation, you can't run Sparc code in emulation mode on x64 gear without an equivalent product or a license to QuickTransit from IBM.
But stop thinking of Moon Macrosystems as a server vendor. At first, think of it like Red Hat, with a precise set of configured systems on which it is installed and partnerships. If customers want Solaris, they surely want it pre-installed and certified on their existing x64 server platforms, not on Sun's "Galaxy" servers - as Sun's own tepid x64 shipments show. (Sun was kicking out some 30,000 x64 servers a quarter, which is a drop in the bucket. Today's total market boasts 2 million or so servers a quarter, most of them being x64 boxes).
That is a short-term hardware plan, just to get going. In the long run, Moon Macrosystems has to offer some kind of differentiation on hardware, assuming that the idea of integrated hardware-software systems or enterprise appliances or whatever you want to call them is real. The important thing about the hardware, the real ASS/500 system, is that it be integrated, like Cisco is doing with the California box in terms of system and storage networking using Fibre Channel over Ethernet for storage. (But Cisco forgot the storage.) Or like HP's Matrix blade platform, which is missing the Fibre Channel over Ethernet.
Moon Macrosystems could just use the Xsigo I/O virtualization appliance and Super Micro whiteboxes and just get around the whole issue. This is Dell's backup plan, and it has the virtue of working on both rack and blade servers. IBM and HP have not done much in terms of virtualizing I/O for rack servers, unless they are Power-based machines in the case of IBM or Itanium-based machines in the case of HP.
You wanna server to chase volumes with Lunaris? You want to be in SMB accounts like HP, Dell, and IBM. Then you need tower servers and blade servers that run on normal wall power. Don't screw this up like Sun has for the past decade.
Now, a brief note about marketing. I think some Moon Macrosystems tattoos placed on celebrity booties would probably do the trick as a publicity stunt. I am thinking JLo for the boys and Hugh Jackman for the girls as a good start. Let the marketing hang out there, people. Just don't talk about being the dot in dot-com, or the Web in Web 2.0 or go on and on about open network systems or some other nonsense. Help people save their asses in this awful economy, and they will help you save your own. All you gotta do to start is find your own with both hands.
One last thing. If by some miracle this actually makes money, save it. Don't ever go public and don't ever let the private equity companies take control. Be happy running a business, employing people, and serving customers. Let that be your reward. All the rest is vapordough anyway. ®