Linux displacing Solaris?

An application that needed a maxed out $200K Sun 450 or HP L class system in the late ninties now runs on a X4200 - for under $10K inclusive of some external disk.
Written by Paul Murphy, Contributor
Grandpa over there, he's just dumb: ran some home brew ERP in something called Progress on a Sun 6500, cost him a million bucks - me, I can do that job using Great Plains and four Xeons for about $10K. What an idiot he was to fall for that SPARC/Solaris stuff - cost his company hundreds of thousands a year, and this went on for like ten years. I mean, what an idiot - and where were his bosses?

Ever want to strangle somebody like that? well, hold the impulse because, except for the anachronisms, the personal insults, and his conclusions, he's got a point.

Okay, he's wrong about the cost. By the time he gets his stuff licensed and in a rack, his employers will have written $70K worth of checks, but realistically that's still only about 7% of what "grandpa" would have expected his bosses to pay and it doesn't take a lot to imagine that today's integrated Great Plains ERP package would be better than most home grown systems.

Even looking at ten year total costs the Windows solution MightyKeen here is advocating is going to beat the mid to late ninties SPARC/Solaris combination by a factor of somewhere between two or three - while delivering software that's almost certainly better. So what's to qvetch?

For openers, Moore's law. There's an awful lot of mid ninties Sun and HP gear that's still cheerfully chugging along earning a living, but its costs then don't bear much resemblence to the cost of comparable hardware and software today. That 6500 produced a pretty fair bang for the buck back then, and Great Plains originally only ran on Macs, but today you'd get comparable bang for considerably less than 5% of the bucks. In fact, if you're paying for support on a 6500 (or an HP K, L, or N series), you can upgrade to a T2000 or X4200 for a couple of months in support fees and save money the rest of the year -while getting better performance and cutting out 80% of the sWaP bill.

What he's fundamentally right about, however, is that application processing needs that used to stretch the limits of million dollar hardware can now be met by little more than entry level gear -and that widely used commercial software has evolved almost as rapidly.

He's wrong, of course, in his implicit belief that Moore's law applies only, or best, to Wintel. None of the x86 boxes that 6500 blew away for cost and performance back then made it much into this century, and today's SPARC and PPC boxes stand in roughly that same relationship to today's x86 products: they cost more, but they work better and will remain viable for business use much longer.

On the other hand, the differential growth rates in computing power and basic application processing needs means that it usually doesn't make sense to upgrade to the current equivelent model if your software load hasn't changed a lot. Replacing, for example, a 300Mhz Sun 450 running Oracle with a modern 490 yields an eight to ten fold improvement in throughput - but if your job has only grown by perhaps 30%, switching to Linux on AMD meets the need at about half the cost.

Thus one of things Moore's law means is that a decision to stick with legacy software while replacing the hardware almost always enables a switch to something that's significantly lower in the "platform" rankings. Thus a mid ninties high end K580 with a 1TB EMC might have cost a million and a half or more and struggled to run an Oracle financials application for a few thousand users - but that same application would now run nicely on a T2000 with a JBOD for less than ten cents on the original dollar. It's true in the middle too: something that needed a maxed out $200K Sun 450 or HP L class system in the late ninties now runs on a X4200 - for under $10K inclusive of some external disk.

Most importantly, it's true for the highest volume, lowest end, systems: a legacy application that needed a mid range Sun box in 1997, now fits on a Xeon running Linux for maybe three cents on the original dollar. Basically what's happened is that the legacy software hasn't changed much, but the hardware has - making Lintel a sensible choice for these kinds of applications and thereby creating an opportunity for people to see Lintel as rapidly replacing Solaris on SPARC.

This, however, is a bit of a mirage. Look at people putting in place enterprise scale network and communications applications, and you don't see much Lintel - because the volume market hardware isn't up to the job and really high end Intel costs significantly more than SPARC. Thus people talk about moving legacy Oracle, SAP Purchasing, or PeopleSoft HR, from SPARC/Solaris to Lintel - but not about swapping their HP N class or Sun 4800s for Lintel so they can continue to run their core XML transformers, document databases, or corporate blogging sites.

So what's really going on is that people who knew what they were doing back in the mid ninties deployed SPARC/Solaris for newer, higher risk, applications at or near the limits of available performance within their range - and that's exactly where it still fits today. In effect, for either time period, its the the newest, most power hungry and business critical applications that define the Solaris/SPARC market.

There is, one key difference: in the late ninties applications that fell off the low end of the SPARC power scale went mostly to NT, now they seem to go mostly to Linux. And that, I think we can all agree, is a good thing.


Editorial standards