Everything is Happening at Once
By Gregp on Mar 31, 2006
So there must be some failure of physics, or entropy gone wild. But from my perspective at Sun, last week was a watershed moment. The public Sun Grid went live, we delivered the source code for the Ultra SPARC T1 processor (nee Niagara) to opensparc.org, and we unveiled Project Darkstar, a breakthrough in massive multiplayer on-line gaming. In the arrow of time, these are all pointing sharply to the future. The debate, it appears, is not whether, but when, computing is a service. If we are receiving any criticism today about our technology strategy, it's that we are being too aggressive about our embrace of the future. That's an enormous change from the barbs of just a few years ago: that we were way too backward looking, holding on to the past computing models. That's the watershed.
Now, anyone remotely connected to high tech R&D knows that you don't just wake up some day, change your view of the future, and then have a bunch of cool things roll out over the next 12 months. It takes years. As I've oft noted, the technology constants change rapidly (faster, smaller, cheaper), but the rate of change for organizing principles and architectures are glacial. The multiprocessing/multithreading concepts behind T1 have been researched for decades, including some excellent workload performance modeling lead by John Busch's team in Sun Labs a half-dozen years ago. It was actually on the basis of that work, circa 2000, that lead to a (very) big bet being taken within Sun to go build a microprocessor that pointed to the future. A design that optimized for the workload of network-scale services. Similarly big bets were taken around Solaris (Zones, ZFS, dtrace, Fire Engine, ...), JES, tools, and AMD's Opteron, just to name a few of them.
The momentum, buzz, excitement that I feel growing among our customers and the broader communities we touch with things like OpenSolaris, OpenSparc and Java, is palpable. It feels like these bets are beginning to convert. Take a look at some folks having some fantastic success with T1, seeing 8x the performance of Itanium on a socket-to-socket comparison
Hey, the last thing we need at Sun is hubris again. That's not the point of this post, at all. It's just that all of those vectors that the 9,000+ engineers at Sun having been hammering away on since the bubble burst are building up to an interesting coherence. More than anything, it's to acknowledge those awesome efforts --- a few of the fruits we saw last week. This is a very special place and a very special time.