Popularity of the current Solaris releases
in [Solaris]
|
Prev: Anyone have a date for the next OpenSolaris ?
Next: UX:lp: ERROR: The LP print service could not read your request
From: yourmommycalled on 2 Dec 2009 23:57 On Dec 2, 1:01 pm, and...(a)cucumber.demon.co.uk (Andrew Gabriel) wrote: > In article <4b15b...(a)212.67.96.135>, > Dave <f...(a)coo.com> writes: > > > > > Andrew Gabriel wrote: > >> In article <Toqdne9RdMmsbo7WnZ2dnUVZ8kqdn...(a)pipex.net>, > >> solx <nos...(a)example.net> writes: > >>> Hi, > > >>> Having worked for a software house, Sun has done themselves no favours > >>> by dropping SPARC workstations. The Sun workstations are all AMD or > >>> Intel based, while servers are SPARC, AMD or Intel. > >>> I worked with developers who wanted SPARC workstations but were given > > >> What did they want them for? > >> If you want to develop for sparc servers, you need to do that on > >> a sparc server. They are now so different from workstations, that > >> if you developed on a workstation, your app would probably run > >> very badly on a modern server. > > >>> AMD64/x86 based workstations running Windows. Hopefully with Solaris 11 > > >> Stick Solaris x86 on it and use it to front-up your sparc server > >> for same look and feel, or even just as an X terminal (or use a > >> SunRay). There's no shortage of ways to achieve the same thing, > >> but developing on a sparc workstation won't any longer give you > >> any feel for how a modern server works - they've moved on too > >> far from simply being big versions of a workstation. > > > Are you referring in particular to the CoolThreads chips? They certainly are a > > totally different animal to the SPARCs in any workstations I've ever used. They > > clearly are capable of high performance if your code can exploit their > > parallelism, or you have a lot of processes, but otherwise they are pretty damm > > slow. I've just about given up trying to develop on a T5240 due to its speed. > > T series and M series - they're very different beasts, but again both > very different from a workstation. For a T5240, you're going to need > somewhere up around 128 runnable threads to get the most from it, > and ideally many more in practice. That's great for throughput > computing such as web serving, and apps which have been designed to > scale to very many threads such as Oracle*, but if you write an app > on a workstation and just move it across without understanding the > differences, you might find you can only use about 1% of the server's > capability. For M-series, you also need a good number of runnable > threads, 16 or more, and you have that combined with cores that each > give you something like 3 times the performance of something like a > V890 core at the same clockspeed. These systems just don't look > anything like big workstations anymore - they've well outgrown that. > I suppose the nearest would be an M3000 with a SunRay. > > *http://www.oracle.com/us/corporate/press/036544 > > -- > Andrew Gabriel > [email address is not usable -- followup in the newsgroup] For some us 16 cores running at 3x the speed of a UltraSparc IV+ would be very nice. Think WRF-DA with 53 vertical layers and 300m horizontal resolution
From: Dave on 3 Dec 2009 06:39 Andrew Gabriel wrote: >> Are you referring in particular to the CoolThreads chips? They certainly are a >> totally different animal to the SPARCs in any workstations I've ever used. They >> clearly are capable of high performance if your code can exploit their >> parallelism, or you have a lot of processes, but otherwise they are pretty damm >> slow. I've just about given up trying to develop on a T5240 due to its speed. > > T series and M series - they're very different beasts, but again both > very different from a workstation. For a T5240, you're going to need > somewhere up around 128 runnable threads to get the most from it, > and ideally many more in practice. That's great for throughput > computing such as web serving, and apps which have been designed to > scale to very many threads such as Oracle*, but if you write an app > on a workstation and just move it across without understanding the > differences, you might find you can only use about 1% of the server's > capability. For M-series, you also need a good number of runnable > threads, 16 or more, and you have that combined with cores that each > give you something like 3 times the performance of something like a > V890 core at the same clockspeed. These systems just don't look > anything like big workstations anymore - they've well outgrown that. > I suppose the nearest would be an M3000 with a SunRay. > > * http://www.oracle.com/us/corporate/press/036544 > I fully accept that programs like Oracle and Apache make good use of machines with CPUs with a low clock speed, but many threads. Unfortunately, computers based on such processors can present problems for many reasons. 1) There may be simply no way to exploit the parallelism when trying to solve a specific problem - some things simply can't be done in parallel. 2) There may be no known algorithm to exploit the parallelism, though one exists. 3) A parallel algorithm may be known, but is significantly more difficult to implement than a serial one. 4) A parallel implementation of a program is possible, but the parts that can't be done in parallel will dominate the CPU time, so the speedup from making use of lots of threads is quite small. 5) Multi-threaded programs are significantly more difficult to debug than serial ones, as failures tend not to be so reproducible. 6) You may be using software written by someone else, which could exploit parallel processors, but which does not. Even if you have access to the source code, you will probably not want to devote the time to convert their program to something which exploits multiple processors. There are probably other reasons too, but those 6 come quickly to my mind. However, I am well aware, that making a CPU run at 1 GHz with 64 threads, is easily practical today, whereas making a CPU run at 64 GHz is not possible today and probably never will be. Dave -- I respectfully request that this message is not archived by companies as unscrupulous as 'Experts Exchange' . In case you are unaware, 'Experts Exchange' take questions posted on the web and try to find idiots stupid enough to pay for the answers, which were posted freely by others. They are leeches.
From: David Combs on 10 Dec 2009 18:58 In article <4b12d316(a)212.67.96.135>, Dave <foo(a)coo.com> wrote: .... > >I bought the other 10 GB from Crucial, and the enterprise grade disks elsewhere, For your site (or thought), what brands of disks qualify as "enterprise" grade? And are those SCSI (iscsi?) or other? Also, is the general understanding that the underlying DISK-HARDWARE for scsi disks is more robust? Thanks! David
From: David Combs on 10 Dec 2009 19:00 In article <x0ocmkz531.fsf(a)Hax.SE>, Thomas Tornblom <thomas(a)Hax.SE> wrote: >Dave <foo(a)coo.com> writes: > >> Canuck57 wrote: >>> Dave wrote: >> >>>> I bought the other 10 GB from Crucial, and the enterprise grade >>>> disks elsewhere, for a hell of a lot less than Sun wanted for them. >>> I will give Sun this, their memory prices are better than IBM or HP, >>> but then their sales are in the tank for UNIX iron. >>> But still, the pricing on Sparc hurts. But at least Sun's stuff >>> works with COTs parts. I too have put memory and disk from COT >>> suppliers. You are at waranty risk but saving $20,000 on a base >>> $20,000 machine is often the benifit. It is that or some sucky >>> MS-Windows boxen from hell. >> >> Does anything beat what HP wanted for RAM to upgrade my HP 1320nw >> Postscript printer? Back in 2006, then wanted �680 to add 128 MB, >> which was far more than I'd paid for printer. It worked out at �5440 >> per GB or $9646 / GB (based on exchange rate in 2006). I ended up pay >> �29.99 for 128 BM, which made HP RAM more than 22x the cost of Crucial. >> > >I just added 128M to the 96M I had in my old HP Color Laserjet 4500 and >paid a whopping $21, plus $5.50 for international shipping :-) > >Thomas Who did you buy it from, HP itself? David
From: David Combs on 10 Dec 2009 19:07
In article <4b12d419(a)212.67.96.135>, Dave <foo(a)coo.com> wrote: >I'm sure its >fastest to develop software on linux, Why? What is that you experience with linux that makes it better for software development than solaris? Thanks, David > >Dave >-- >I respectfully request that this message is not archived by companies as >unscrupulous as 'Experts Exchange' . In case you are unaware, >'Experts Exchange' take questions posted on the web and try to find >idiots stupid enough to pay for the answers, which were posted freely >by others. They are leeches. |