Bryan J. Smith b.j.smith at ieee.org wrote:>Not only that, but people forget that SPARC is not sold by just Sun. >SPARC is an IEEE standard licensed under "fair and non-discriminatory" >terms. The SPARC ISA and most architectural details are freely >available.Yep - but what is your point with bringing that up?>The _majority_ of my Solaris/SPARC experience in more recent years has >been on Fujitsu/HAL solutions. I not only prefer the Fujitsu/HAL >products over Sun, I not only prefer Fujitsu's "openness" to support >SPARC solutions with non-Fijitsu networking, storage, etc... attached >(whereas Sun likes 100% Sun equipment from a support perspective)Sun will very willingly support all certified solutions (including EMC, netApp or Hitachi for storage) as well as provide some support with uncertified hardware for us... no issues...> but >Fujitsu has taken over packaging design and fabrication of SPARCs for >Sun itself. I.e., Sun used to design their own SPARC packages and TI >was their foundary, but Fujitsu has almost always designed their own >SPARC modules/ daughtercards, and fabbed them themselves.Yep - but how is that a benefit if your stuff works? NVidia is fabless i.e. and you surely don't want to tell me they aren't good at what they do... Having your own fab or not has very little influence...>Sun even >sells Fujitsu PRIMEPOWER products on their site now.They don't - they had info for it a while ago but even that has been removed.>sun4u (UltraSPARC) is planned through UltraSPARC VUS-V has been canceled...>, with UltraSPARC III >(and lightweight IIIi) and IV available today. The 1-2 way UltraSPARC >IIIi is more comparable to the cost/design of the Pentium III and IV, as >well as the Xeon.US-IIIi is up to 4 way...>The UltraSPARC III and IV, and their platform, are >_clearly_superior_ to even OEM proprietary NUMA Xeon implementationsYep>(you really have to start looking at proprietary IA-64/Itanium II to get >a good comparison). > >But the design of the Opteron is why Sun is moving forth with it's move. >The typical UltraSPARC NUMA/SBUS server architecture really doesn't >offer much over Opteron in a 2-8 way.Sbus ???? Now you're talking US-I/II again - and that is absolutely no match for an opteron box...>And to be honest, Solaris is >probably the most "mature" OS for Opteron 200/800 right now, given its >experience on partial mesh interconnected systems.Based on what? The Implementation of Solaris for AMD dates back to SE b68 or 69... that's mid 2004... Compare that to an OS that already ran on the software emulator before silicon was even out there... Solaris 10 on AMD64 is definitely less stable that Centos 4 on the same v40z... Running the same apps we had 0 crashes with Linux vs. 2 explained and patched, 1 explained still unpatched and 1 unexplained crash on Solaris10...>Linux will get >there, but it takes time and experience on a platform like SPARC >(Windows never well, not even thanx to SGI's past, although brief, >donations).Why would you need to run on sparc to produce a stable OS for an opteron box? Peter.
Bryan J. Smith <b.j.smith@ieee.org>
2005-Jun-30 23:55 UTC
[CentOS] [OT] SPARC platforms -- WAS: Hot swap CPU
From: Peter Arremann <loony at loonybin.org>> Sun will very willingly support all certified solutions (including EMC, netApp > or Hitachi for storage) as well as provide some support with uncertified > hardware for us... no issues...You work in a Fortune 5, they will do anything you tell them. I've been there before, and that is a very nice position to hold. But not when you work for a 50 engineer semiconductor startup. Sun really pissed us off regularly, so we stuck with Fujistu.> Yep - but how is that a benefit if your stuff works? NVidia is fabless i.e. > and you surely don't want to tell me they aren't good at what they do... > Having your own fab or not has very little influence...I was just saying that anyone who only wants to go to Sun because they are behind SPARC might consider Fujitsu, who actually fabs the chips. I've walked into many shops who have had issues with Sun's support and they never knew about Fujitsu as an option. That's all.> Sbus ???? Now you're talking US-I/II again - and that is absolutely > no match for an opteron box...Sorry, I meant NUMA/UPA -- my bad.> Based on what? The Implementation of Solaris for AMD dates back to SE b68 or > 69... that's mid 2004... Compare that to an OS that already ran on the > software emulator before silicon was even out there...Solaris was _also_ running on the x86-64 emulator too. ;-> AMD and Sun were in negotiations before the first Opteron came out.> Solaris 10 on AMD64 is definitely less stable that Centos 4 on the same v40z... > Running the same apps we had 0 crashes with Linux vs. 2 explained and > patched, 1 explained still unpatched and 1 unexplained crash on Solaris10...I meant the design: "given its experience on partial mesh interconnected systems" I thought by putting that in the _same_statement_ you'd see what I meant by "mature." Sorry, I assumed wrong. ;->> Why would you need to run on sparc to produce a stable OS for an opteron box?Re-read my full statement ... 'Solaris is probably the most "mature" OS for Opteron 200/800 right now, given its experience on partial mesh interconnected systems.' -- Bryan J. Smith mailto:b.j.smith at ieee.org
Bryan J. Smith <b.j.smith@ieee.org>
2005-Jul-01 00:25 UTC
[CentOS] [OT] SPARC platforms -- WAS: Hot swap CPU
From: Peter Arremann <loony at loonybin.org>> US-V has been canceled...By Sun, yes. They believe they will transition to Opteron sooner than they thought. All I said was that it had been "planned through" US-V, including the design work done. My bet is that the engineers had finished the core design, even if not fully optimized, and that's when they were canned. It's still an option in the back of Sun's pocket. [ I.e., the core design occurs _years_ before products become available. E.g., even the latest Opterons are based on the 1997 core design. And until the Pentium 4 came out (just a refit), the Pentium 3 was still based on the 1994 Pentium Pro core. ] Fujitsu has yet to weigh in and possibly pick up the design. They have been designing SPARC modules (better ones IMHO) for a long time. All it takes is 1 customer who needs SPARC binary compatibility. There was an equal battle over the Alpha 364, with OpenVMS the driver. Especially since OpenVMS/Itanium was _slower_ (than an Alpha 264 from 3 years earlier no less!). But once [quality] emulators became available, that killed it. I don't think you'll see the same with SPARC, but I could be wrong. If the need by just 1 major user is there, you can be sure Fujitsu will come out with a PRIMEPower upgrade that uses the design. I never said Sun would release an UltraSPARC V, I just said it has been planned (i.e., designed). Sun has made it known that as of right now, UltraSPARC IV is the "end-of-the-line" for Solaris, and given how Solaris is working on Opteron, I don't blame them. But HP is _still_ selling Alpha 264s because of OpenVMS. ;-> A platform they have tried to kill off again and again and again.> US-IIIi is up to 4 way...I _never_ said it wasn't. I just said the 1-2 ways are comparable to a typical PC design**. But after 2-way, it's much better to go with a US-III or US-IV, kinda like it's much better to go with an Opteron instead of a Xeon beyond 2-way**. ;-> -- Bryan **NOTE: This is a recurring theme. You seem fixated on the assumption that a system that uses a single-point-of-contention design (namely the Intel MCH aka "northbridge") approach would _always_ be faster. You seem to lack the understanding beyond the real design contraints of a partial mesh interconnect on a real system interconnect (and not just a peripheral interconnect hacked on as a system interconnect). You have degraded most of my commentary as a "that's a chipset issue." You seem to have a very limited understanding outside of the traditional PC CPU-to-memory or CPU-to-I/O via a "chipset" and that's why you keeping missing my points on Opteron, UltraSPARC, etc... Not to mention this thread was just about finding a solution with hot-swappable components. I left it at that _until_ you wanted to assert all sorts of things (using "builds" as a benchmark for servers, even computational ones). -- Bryan J. Smith mailto:b.j.smith at ieee.org