On 27 May 2012, at 04:34, Garrett D'Amore wrote: <snip> > In fact, I'd wager that the vast majority of commercially interesting users > of Solaris usually just *reinstall* when going from one version to the next. > The "upgrade" even before the major changes in OpenSolaris/SXCE/S11 was > always so painful, so unlikely to work properly, and so generally > *untrusted*, that almost nobody in *production* environments ever just did a > an OS "upgrade". It was easier to just fresh install and move the > applications & data over.
Absolutely. Liveupgrade took some of the pain away, but I agree just reinstalling apps and moving the data/config across when things change substantially is often safest and helps avoid any nasty surprises. > Note that the key issue here is *applications*. There is another solution to > that problem. Its called virtualization. And in S11 they have S8 and S10 > zone support. In illumos we have kvm, which means we can host *any* older > Solaris (x86 only), I think. That will provide a migration path, retaining > the crufty concerning issues (compatibility of applications, APIs, and most > importantly ABIs) while still letting folks move onto newer hardware. > S'noracle realized this and its part of why they abandoned so much of their > legacy -- because they could just port that legacy into an S10 or an S8 > branded zone. :-) Exactly - if you need old, use old. Why put new on old? Use new! <snip> >> SPARC is a technology owned by Oracle. The only SPARC boxes you can buy on >> eBay are relics, and I'd wager my Samsung Galaxy S2 phone with it's dual >> core 1GHz ARM chip can outperform all of them. Power is now incredibly >> expensive, and the economics of running old equipment make no sense what so >> ever. If you buy an old SPARC box on ebay for $500, it's going to cost you >> more than that per year to run it. > > I highly doubt your 1GHz dual core ARM can match a 64-way 450 MHz E10K, > although its probably a lot closer than I'd like to admit :-) And that's a > system that is very very old indeed. There are some powerful (especially in > their day) SPARC systems still out there. There are, and there is the legacy application argument. But these machines are machines that use kilowatts - an E10k can use 13kW. In London datacenter power cost that's around $6k/month. For one months power budget you can get an Intel box that outperforms it quite comfortably. There are obviously newer, less power hungry boxes, but I'd bet this "1 months power budget gets you a comparable Intel box" rule of thumb probably holds true for most of them! In the $dayjob we just replaced a large number of machines that are only just 2.5 years old because they had 2 x Intel E54xx CPUs in (8 cores total, so not useless) and each machine used 1.2A of power. We can now get 12 cores per machine and they only use 0.8A. So economically it was a no brainer. And nobody wants these boxes (Sun Fire x4150) on eBay now despite pricing them at a fifth of what we paid for them. What makes hardware replacement so incredibly economical and easy is when you stick everything in a virtualisation layer, because moving your applications becomes trivial and the manpower cost becomes negligible. It just becomes a very simple mathamatical calculation and there is always a tipping point where it makes sense to replace that old hardware. IT depreciation assumes a 3 year lifecycle anyway. So purely from economics alone, SPARC just doesn't make sense to me, unless you don't like polar bears (CO2) or aren't paying for the power out of your budget directly. If someone disagrees and can demonstrate the economics then I'm all ears. >From the legacy applications standpoint, again for power reasons you're better >off buying a modern SPARC box and virtualising your workloads, and at that >point you're better off going for Solaris 10 or 11 because they are more >likely to support your legacy application, either natively or in a branded >zone. > The problem with SPARC is that they haven't kept up with current Intel > hardware, and that some of the platform specific pieces of support code were > never opened. None of them are particularly hard to reproduce, but who wants > to spend a bunch of time working on code for systems that nobody can afford > to power on any more? Show me someone who *really* is committed to spending > lots of money to keep those systems running - not just running, but updatable > to a newer OS -- (both in terms of power/equipment, and int terms of dollars > to pay for engineering work), and I'm sure we can get it going again. Heck, > I'll even offer to help if there really truly is some evidence that this is a > worthwhile venture and not just some hobbyist hackery. (Don't get me wrong, > hobbyist hackery is great -- but I've got far too many critical forward > looking projects to worry about garage retro computing projects -- that's the > demesne of NetBSD. :-) > > In fact, developer-council met and one of the topics briefly discussed was > SPARC. Most of us hate the trouble it is to keep it around, but we agree > that we need it. We need it because it keeps us honest (SPARC is both an > alternate architecture *and* a big-endian one), and because the backlash that > would come about if we yanked it would be -- unfortunate. So we're keeping > it alive, if it can truly be considered to be alive even now. Still, none of > the distros have taken up the work to build it, so all we have is kernel bits > that we *think* work (at least they did the last time I tried, which was a > couple of months ago.) Yes those are good reasons for keeping it around and the only reason we (OpenIndiana) didn't do a SPARC build was because it doubles the release engineering effort - an area we struggle hard enough with as it is due to the tediousness of it. Dildos supports SPARC (I believe igork has a working build) and I believe Illumian will too. I would welcome any SPARC enthusiasts to volunteer their time to help us get SPARC support added to OI. Work has been done and we were donated a T1k to assist with this, it just needs to be completed. >> You're far better off buying a high density modern intel box and then use >> virtualisation to consolidate workloads. You just cannot beat the economics. >> There is just no viable reason for using SPARC, except perhaps to run legacy >> software where no binaries exist for x86. > > *Some* of that legacy application base is *hugely* important. But its also > the case that those apps aren't going to be looking to move to new operating > systems any time soon. They just want to keep running. They might want to > do so on new hardware. A really good SPARC emulation running on x86 might be > a compelling offering for these guys. I'm not aware of any that are robust > enough for production right now. There probably *is* a niche, yet valuable, > market for such a solution. (Could be a good startup for a very small team > of dedicated virtualization engineers with SPARC backgrounds. I think some > of the initial work is already done in qemu.) QEMU SPARC support is interesting, and modern CPUs keep getting faster. I wonder if a single SPARC thread under QEMU would be comparable to a single Intel core under QEMU :-) It won't be long until Intel have boxes with as many cores as older T series has threads. >> Since future SPARC development is now under the stewardship of Oracle, it's >> now a proprietary legacy CPU architecture. > > Does Oracle own "SPARC", or just the only company doing SPARC designs? (Have > Fujitsu pulled out entirely of the SPARC space?) As best I know that alliance is still going but when I spoke to Fujitsu about sponsorship for OpenIndiana, it seemed to get passed around quite a bit, and then we were politely told due to their relationship with Oracle it would be a conflict of interest. www.opensparc.net hasn't been updated since 2010, although it does have some Oracle content (the logo at the bottom has been updated to an Oracle one). But it doesn't look like they are keeping it up. While there's source for T1 and T2, there's certainly no source for T3 or T4. So although Fujitsu is still pushing SPARC, I fear they are tainted through association. Oracle are litigious and like to protect any IP they view as valuable. Given their commitment to SPARC and their current direction, I suspect the new SPARC CPUs are about as open as Solaris 11. >> If you buy a new SPARC box, you're giving your money to Oracle and at that >> point you may as well run Solaris 10 or 11 on it. > > That's certainly true. I can't imagine why you'd fork over the premiums to > Oracle for their hardware unless you needed their support guarantees that > certainly only come with their OS. > >> The illumos developers have commercial interests wedded on x86, > > Not all of us are wedded to x86. But the lack of any viable alternatives > leaves us there. :-) I'd love to see a high end MIPS or ARM port. I think > the problem is that all the chip guys but Intel and AMD are focussed almost > entirely on the consumer device space now. (Frankly, its the only way to > compete if your a chip mfg. The Windows monopoly still dominates the desktop > space. Client devices have a much more level playing field.) I know of someone who is working on resurrecting ARM - I don't know if he wants it public but he is making progress there. Drop me a line off-list if you're interested and I'll put you in touch. >> and since the project has no access to modern SPARC equipment like M3000s or >> T4s, nor access to the technical specs, there's no real reason for the >> illumos developers to expend valuable time and energy on developing for >> SPARC. > > We've been offered access to M3000 class hardware. (Not T4.) No specs, and > without help from Fujitsu the M-series work would probably be dead in the > water. I'm guessing that M3000 wasn't offered by Fujitsu! Or if it was, well done on speaking to the right person! >> So at this point, as far as I'm concerned the people who want SPARC are >> hobbyists interested in running old equipment, much like how I like to get >> out my Commodore 64 and BBC Micro, or they're lunatics with no grip on >> reality. > > That's a bit ... over the top. But yeah, IMO there is little commercial > reason to develop a SPARC distro. Frankly, the lack of one is probably the > single most compelling evidence in support of the argument. My whole email was kind of over the top ;-) I couldn't resist wading in, mainly because I felt the SPARC/SVR4 folk need some of this: http://www.youtube.com/watch?v=hLxQQW-rtDY > However, I've got some SPARC experience and expertise, and if someone really > wants to do the work, I'm happy to help mentor individuals working on the low > level pieces that are still missing. (Please be a reasonably experienced > kernel guy though. This isn't a project to cut your kernel teeth on -- I > don't have time to handhold a newbie.) > >> Secondly, SVR4 is yet another dead end technology. SVR4 packaging was crap >> back when it was widely in use, and it's still crap now. OpenIndiana and >> OmniOS use IPS, Nexenta/Illumian use deb, and SmartOS dispensed with >> packaging entirely. If you have SVR4 packages, rinse them through pkgsend >> which will publish them to an IPS repo and problem solved. >> >> When people talk to me about SVR4, I imagine things like this: >> >> http://en.wikipedia.org/wiki/File:SunOS_4.1.1_P1270750.jpg > > Hah, that's SunOS 4.1.1. It was most definitely *not* SVR4, but BSD based. > I remember installing that OS from tape. :-) Haha, good point! I was 8 years old when that came out, so unfortunately never had the opportunity to try it out :-) >> Sure, there are people out there with fetishes for cassette tapes and vinyl >> records, but they, like the SVR4 fetishists, are in the minority. The >> majority of people just want to get work done, and IPS lets you get more >> work done faster. > > I disagree, except that IPS is what we *have* today. We don't have SVR4. > This is not intrinsically, IMO, to the vast superiority of IPS, but due to > the fact that Oracle made the change, and almost nobody sane has been willing > to do the work to revive SVR4. Truth is, we don't need a packaging war. > Btw, in *my* commercial product there is *no* packaging whatsoever. The > "distro" is a self contained ISO. But, I'm not doing a general purpose > operating system either. :-) Exactly - packaging just simply isn't all that interesting, and if you can dispense with it entirely then all the better. Although project lead of OI, we chose to roll out SmartOS, because the way it works matches our workload (cloud computing) perfectly. It netboots into RAM and the zones are sparse. We also get to have our cake and eat it because we deploy IPS in the zones under an image rooted at /ec. Managing OS updates one way (update netboot image and reboot) and application updates another (pkg update) works exceptionally well in a cloud computing environment. We're very very happy with our environment and workflow and I'm over the moon that the base OS isn't managed via any kind of packaging. >> If you're pining for SVR4, put your money where your mouth is and fork. I >> can guarantee you that the unwashed masses will give your 80s throwback >> distro a wide berth. > > Its been done already. Schillix. Have fun. :-) That's such a good point! >> As project lead I can tell you that we won't be doing SPARC unless someone >> comes along and takes ownership of that project, and we definitely won't >> ever be doing SVR4 packaging as long as I'm project lead. > > Please be clear, you *only* speak for OpenIndiana and most *definitely* not > for illumos. You are not the project lead for illumos. Oh absolutely, I am only speaking here for OpenIndiana. > (If there is one at all for illumos, it is me. But as part of the most > recent developer council meeting I think basically I agreed that such a title > is fairly empty since any meaningful decisions are made by the developer or > admin councils.) > > Other distros can do what they like; we will welcome a reasonable SVR4 port > in illumos. (Which isn't to say we will necessarily *use* it.) I would also invite anyone who wanted to do so to base it off illumos-userland, as the more people we have collaborating on porting great software to illumos the better! Alasdair ------------------------------------------- illumos-discuss Archives: https://www.listbox.com/member/archive/182180/=now RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be Modify Your Subscription: https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4 Powered by Listbox: http://www.listbox.com
