Quite a launch today. The Galaxy launch had its moments, too – it got people ready to treat Sun seriously as an x86 systems vendor, and the machines were quite nice. But, at the end of the day, they were 1U and 2U Opteron servers, and those aren’t exactly hard to find.
Today’s launch, however, is quite a different story: three significantly different products all landing at once, and all quite different from anything that you can get from other vendors. Galaxy 4 (a.k.a Sun Fire x4600) is, I suppose, the most straightforward of them: 8 Opteron sockets in a 4U box. Dual core now; when quad core comes out, you’ll be able to slide out the two-core CPU modules and slide in the four-core modules. 64GB of memory now, 128GB later. If you need a lot of compute power, it should do the trick; we sold a bunch of prerelease units (plus some Thumpers) to Tokyo Tech, and now it’s the fastest supercomputer in Asia.
The one that I understand the least is Andromeda (Sun Blade 8000). I’ve simply never worked in the sort of data center environment where blades’ virtues are at a premium, so I can’t say I entirely understand the pros (and cons, I suppose, but I don’t really know what they might be, other than perhaps its 19U height) of this versus our competitors’ designs. As a good object oriented programmer, though, the separation of I/O and processing elements sounds pretty interesting, with both sides independently replaceable and manageable. And they’ve done a lot of work towards increasing the lifespan of the chassis and various design elements, so it shouldn’t go obsolete on you soon: it should live up to the blade promise of simplified management combined with ease of growth. And once we, say, release Niagara blades for it (I haven’t heard any details about that, but it’s got to be coming), the ecosystem will get even richer. (And all you fan fetishists will enjoy standing behind it and feeling the wind blow through your hair.)
But by far my favorite is Thumper. (Sun Fire x4500.) I’ve been using prototype versions of the hardware for the last two years, and I still love to take off the cover and look at all those disks. (Jonathan put up a picture.) 24TB of storage in a 4U box; to put it another way, if you take four racks full of Thumpers, you have a petabyte of data. That is a lot of storage in not very much space.
And you also get a couple of dual-core Opterons in each box. (Or: you have 160 quite powerful cores to comb through your petabyte of data.) Next to Galaxy 4’s 8 sockets, a mere two-socket server doesn’t sound like so much, but let me assure you that it’s quite a lot of compute power to manage, mine, and process that data. I’m not creative enough to envision all the uses that the world will find for that, but early reports are already surprising me with their ideas, and we’ve certainly had a lot of fun playing with them.
Post Revisions:
There are no revisions for this post.
Posts
For whatever it’s worth, my eyes glaze over a bit with this kind of stuff. From my point of view, there’s sort of an endless progression of not-particularly distinct boxes that parade in and out of colo facilities everywhere, so it’s hard for me to see more than just a new point on the big fuzzy curve whose general direction isn’t even remarkable any more.
But I’ve been excited about stuff before that seemed pretty obscure to most people (proof-carrying code?); and I’d probably have said it was because I saw newly possible end results that they didn’t. That is, points on the curve *are* remarkable when suddenly I can do something specific with it that I’d been dreaming of.
So, what does this mean in terms of applications? Does it bring some particular kind of task within a particular audience’s reach? Who has 24TB of data? What sort of stuff did you do with the prototypes that you particularly enjoyed?
7/14/2006 @ 2:19 pm
I sympathize with the ‘eyes glaze over’ part – that’s kind of the way I feel about Galaxy 4. I mean, I’m glad that we’re doing an incredibly powerful relatively compact server, but it’s not like I’ve never seen fast multi-core servers before. The difference to me when looking at Thumper (or when carrying it, for that matter) is that I’ve never opened up a box and seen anything like the density of disks in this one.
I wish I could talk more about what we’re doing with it; it is very frustrating. I hope we will have a product announcement one of these months, but I’m not quite sure when. Or what I should be doing to hasten the day that we will have an announcement, for that matter – I have an idea of what to do if I want to, say, implement a particular feature as quickly and robustly as possible, but I’m at more of a loss when trying to figure out how to get from the state “customers aren’t sufficiently convinced that our product can do what they want better than competitors’ products” to “customers are convinced that …”. So many variables going on there – picking the features to implement, working to define the features when nobody is sure what the best behavior is, various business-related negotiations, … It was a lot easier when I was just hacking GDB either for fun or to implement features that I needed in my own debugging.
7/14/2006 @ 2:50 pm