One other fun thing about the Fowler-DHH interview that I mentioned recently: about 37 minutes into the podcast, the conversation turns to large companies and their involvement in open source in general, Ruby in particular. They initially start off dubious about the concept, with Microsoft as their example, which made me wonder “hey, what about my employer?” And, right on cue, DHH brings up Sun: Rails may have started partly as a rebellion against some aspects of Java, but that doesn’t mean that the Ruby and Rails communities don’t appreciate JRuby. (Which just released version 1.0, yay!)
Not that it was a big Sun love-in, or anything: the discussion mostly went back to Microsoft bashing, spurred by their recent “253 patents” threat. That discussion was interesting from a Sun insider point of view, too: they speculate that some important people at Microsoft were really annoyed at those threats, but Microsoft is too disorganized to have a strategy in this regard. Which I can’t imagine happening at Sun right now: I’ve been really impressed recently at how coherent we’ve been on our strategy. Sure, we have a Chief Open Source Officer who’s doing a great job, but the open source support goes all the way up and back down. (I get the feeling that Sun wasn’t so coherent in the years before I showed up; don’t have direct experience, though, so I could be wrong.)
We still have a way to go, admittedly. There was a thread half a year or so ago on a red-bean mailing list in which some of the participants were annoyed at the open source talk coming out of Sun, feeling that our talk outpaced our performance, so clearly we have some work to do to build up our credibility. But I’m hearing more and more kind words from the community these days. Which speaks to the openness of the open source community, too: I’ve been very impressed in particular by how we’ve been welcomed by the free Java community. Clearly there are a lot of people there who just want the best free Java implementation that they can get, and are willing to accept help from wherever they can get it, not letting inertia and preconceptions get in their way. (Which has been evident for some time: I was floored by Tom Tromey’s proposal a year and a half ago saying that he thought the best thing for gcj was to throw away code that he’d spent a lot of time working on and replace it with the Eclipse compiler.)
And then I see Eben Moglen saying something along the lines of “In fact Sun could, moving forward, in principle be seen as the commercial avatar of the FSF.” Which is just amazing, and it’s not the only piece of support the FSF have given us recently. (Incidentally, I didn’t realize that Balaji was an open source supporter; he’s on the StreamStar team! And I see some other StreamStar programmers in that picture, too! Alas, StreamStar is not currently open sourced; I’d be happy to present my point of view on that should anybody care, but it has a lot more to do with the extremely small size of the team than with any active desire to keep the software proprietary, and I can easily imagine us open sourcing it in the future.)
I suppose I might as well get my Sun boosterism out of the way all at once, on the theory that people who are turned off by that stopped reading a couple of paragraphs ago. So:
- Go Constellation! I’m far from a blade expert, but I get the impression that it’s quite a system: a good job of balancing processor, memory, I/O bandwidth (not sure how much I can talk about it, but we have some very impressive evidence to that end coming up soon), and a good job of avoiding proprietary lock-in. (E.g. using PCI express for I/O cards instead of inventing an interface that other people would have to license.) And while Opteron chips have served us well, I’m really glad to see a product with Intel chips launched, and that we can mix and match with Niagara as well. My favorite part, though, is the rack-sized blade chassis that you can see about 15 minutes into the third part of the launch video: I just like big honking products these days.
- ZFS has been out for a little while now, but I’m glad to see that people are continuing to take notice of it. Here’s the most recent example that I’ve seen. Certainly none of my disk experiences over the last half year have given me any reason to doubt the excellence of ZFS’s ideas…
Post Revisions:
There are no revisions for this post.
Posts
I think what enhanced Sun’s credibility for me was the choice of the plain old GPL for the Java stuff, not some custom license. I mean, Sun says “GPL”, and we say, “… Okay.” :)
Your Sun boosterism never makes me stop reading. But I am utterly nonplussed by your excitement over big honking products. I mean, congrats to Sun, but from the point of view of a non-stockholder, isn’t somebody going to turn the dial to eleven? There are so many dials to turn; it’s not like Constellation’s choice of dial is particularly interesting. You’re a guy with good taste; why is this exciting?
6/14/2007 @ 1:23 am
Yeah, I was surprised and impressed by the GPL v2 choice for Java, too. I suspect that at least some of the reasons why we went for CDDL over GPLv2 for Solaris were good ones, and I also think that many in the Linux community seriously underestimate the amount of effort needed to open source a 30 year old code base like Solaris. But the fact remains that CDDL was a custom license that limited interoperability with Linux, and you could easily come up with reasons why that might be good for us for competitive reasons. Whereas with Java, we can throw together a lot more of the free stuff at once now.
To be honest, I’m not sure why I like the big honking products, either. Maybe it’s that, in the past, I haven’t spent so much time in server farms, so the scale of some of these things is new to me? In particular, I’d never seen anything like Thumper before: you open one of them up, or at least I do, you see all those disks jammed together, and you say wow. Somehow it was a more visceral statement than I’d seen before.
Also, again for me, some of this wasn’t just about turning the dials up to eleven: more like turning the dials up to 20. Three or four years ago, when we started getting the first prototype Thumpers and StreamSwitches, I was used to having servers with tens to hundreds of gigabytes of storage, and a gig or two of memory. All of a sudden, I was working with a single server with tens of terabytes of storage (and you could hook them up to get hundreds of terabytes), and another server with a terabyte of memory. So, all of a sudden, the capacity of servers that I was working with jumped by about a factor of a thousand. We’ve all gone through that before, but I’m used to its taking a decade or so for that to happen, not for it to happen all at once. (The amount of memory and storage on my personal machines hasn’t expanded by that much in the 9 years that I’ve lived in California.) Again, that says as much about the environment that I was coming from as anything else – I assume that there were machines around with that much disk storage and memory, though I’m not sure they were quite that small – but it had an impact on me.
Admittedly, Constellation, even in its full-rack version, doesn’t have the same impact for me, and there’s no particular reason why it should for anybody else. It’s a big step towards helping us re-enter the HPC market in a serious way (which we started a year ago, but we need to keep the momentum going), and I’m sure there are corporate customers who will find it fits their needs very well, but while that’s great for Sun, it’s not like people wouldn’t have been able to build supercomputers without our products. I actually like the openness and modularity of the architecture more than I like the scale; I’ve probably been sucked in by marketing hype (I haven’t looked into the IBM and HP blade products at all closely), but I get the impression that it’s a lot easier to mix and match modules (at least I/O ones) on our blades than it is on our competitors, and that they’re going in for proprietary lockin in ways that we aren’t. I’m sure they’ll come around, but it’s nice to be in the lead in hardware openness in this instance. (If we in fact are – see the marketing hype disclaimer above.)
6/14/2007 @ 8:19 am