[fsug-tvm] Re: Dual Cores
Hi, I didn't want to revisit this thread again (sorry!). But coincidentally, this interesting story just broke on Slashdot: Story: Windows and Linux Not Well Prepared For Multicore Chips Original source: http://www.infoworld.com/article/09/03/20/Multicore_chips_pose_next_big_challenge_for_industry_1.html Slashdot article: http://developers.slashdot.org/article.pl?sid=09/03/22/193205 In particular, this comment is noteworthy: http://developers.slashdot.org/comments.pl?sid=1171173&cid=27291945 Regards, Gokul Das --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
BINNY THOMAS wrote: > BeOS, do you mean Berkley OS? No. BeOS was developed by Be Inc in early 90s. They operated from California till it was dissolved in 2001. The name Be is not related to Berkeley and has an interesting story behind it. The story is available on wikipedia page on Be Inc. Regards, Gokul Das --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Hypertransport... hmmm.. interesting! Seems it is an open standard. ശ്യാം ചേട്ടാ, thanks for the info on FSB replacements. That one goes into my collection of open specs ;). Regards, Gokul Das --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
BeOS, do you mean Berkley OS? On Sun, Mar 22, 2009 at 5:00 PM, Visakh wrote: > > Syam Krishnan wrote: > > Even before multicore procesors became common, there were > multi-processor systems. So operating systems have been handling multiple > processors for a long time. So you need not worry about that. > > Just some off-topic info related to this. Back in the early 90s > there was one proprietary Desktop-specific OS which did an amazing job > on SMPs. It was BeOS. Back when multicore processors were not even > imagined for home use, even before windoze became a multiuser system, > it could run on upto 8 separate processors. We could even control how > many processors it used. It did back then what many OSs struggle to do > even today- play a dozen media files without jerking. Unfortunately it > died a premature death at the hands of an 'all powerful monopolistic > corporation'. > > Now there is an Open source project which strives to bring BeOS back > to life: the Haiku project. Their emphasis is on efficiency and > integrity of the system. They don't even use the X server. The project > seems to be undergoing some interesting development now. > > Another OS which has great SMP support is Solaris. I wrote a school > assignment on OSs- long before OpenSolaris was released. From then, I > remember that it could support a 64 node SMP. That is impressive since > SMPs are extremely hard to scale beyond 8 due to severe bus > bottlenecks (NUMA is used after that). Now it seems to have risen to > 128 nodes! So if you are looking for a large sever OS, openSolaris may > be the answer. > > And now info that really does matter. WinXP (Pro) seems to support > only 2 cores - not useful on a quad core (please note!) . And Linux > may be able to support 32 node SMPs. (Makes me wonder what the > condition of the bus will be like! ) > > Regards, > Gokul Das > > > --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Nice info Gokul.. I had read in a very old edition of PCQuest about this 'multiple videos without jerking' with BeOS. And regarding processor bus, there's AMD's HyperTransport and Intel's QuickPath Interconnect (QPI - debut with the new Core i7). Syam --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Syam Krishnan wrote: > Even before multicore procesors became common, there were multi-processor > systems. So operating systems have been handling multiple processors for a > long time. So you need not worry about that. Just some off-topic info related to this. Back in the early 90s there was one proprietary Desktop-specific OS which did an amazing job on SMPs. It was BeOS. Back when multicore processors were not even imagined for home use, even before windoze became a multiuser system, it could run on upto 8 separate processors. We could even control how many processors it used. It did back then what many OSs struggle to do even today- play a dozen media files without jerking. Unfortunately it died a premature death at the hands of an 'all powerful monopolistic corporation'. Now there is an Open source project which strives to bring BeOS back to life: the Haiku project. Their emphasis is on efficiency and integrity of the system. They don't even use the X server. The project seems to be undergoing some interesting development now. Another OS which has great SMP support is Solaris. I wrote a school assignment on OSs- long before OpenSolaris was released. From then, I remember that it could support a 64 node SMP. That is impressive since SMPs are extremely hard to scale beyond 8 due to severe bus bottlenecks (NUMA is used after that). Now it seems to have risen to 128 nodes! So if you are looking for a large sever OS, openSolaris may be the answer. And now info that really does matter. WinXP (Pro) seems to support only 2 cores - not useful on a quad core (please note!) . And Linux may be able to support 32 node SMPs. (Makes me wonder what the condition of the bus will be like! ) Regards, Gokul Das --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
BINNY THOMAS wrote: Well the reason I am interested it in this topic is because I installed Linux on a machine with 4 cores and the system monitor showed 4 cpus with their utilization graphs. I hadn't seen this windows Task manager before so I regarded it as a plus point for Linux until I saw the same in windows. So i decided to know more about it.I have known about the under utilisation of all cores by the os because most oses are written for only a single processor(not including the gpu). The OS can make use of the multiple cores. Even before multicore procesors became common, there were multi-processor systems. So operating systems have been handling multiple processors for a long time. So you need not worry about that. Now, when it comes to applications, as Gokul said, it depends on how they are written. Multi-threaded applications (with reasonably independent threads) benefit from having multiple processors, while I/O bound sequential processing apps may not benefit much. Here's a tip.. The GNU make command has a command-line argument -j to specify the number of jobs. Setting this number to a value slightly more than the number of cores is supposed to give better performance while building. regards, Syam --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Well the reason I am interested it in this topic is because I installed Linux on a machine with 4 cores and the system monitor showed 4 cpus with their utilization graphs. I hadn't seen this windows Task manager before so I regarded it as a plus point for Linux until I saw the same in windows[?]. So i decided to know more about it.I have known about the under utilisation of all cores by the os because most oses are written for only a single processor(not including the gpu). Thanks On Sat, Mar 21, 2009 at 10:11 PM, Sunil Thomas Thonikuzhiyil < vu2...@gmail.com> wrote: > Oops. My reply caused a lot of confusion. I was referring to kernel > compilation as a possible use of multi core for a power user. > See this > https://help.ubuntu.com/community/Kernel/Compile > Sunil > > > On Sat, Mar 21, 2009 at 2:46 PM, Sunil Thomas Thonikuzhiyil < > vu2...@gmail.com> wrote: > >> Depending on what you do. If you compile kernel you can speed up . >> >> Sunil >> >> >> On Sat, Mar 21, 2009 at 1:52 PM, BINNY THOMAS wrote: >> >>> I want to know about the multiprocessor support of linux i.e. Dual >>> cores,Core 2 Duo,Quad cores etc. Can it effectively utilise all the cores? >>> >>> Thanks >>> >>> >>> >> >> >> -- >> Sunil T T >> Assistant Professor >> Dept. of Electronics >> College of Engineering >> Attingal Pin 695 101 >> http://brainstorms.in >> >> > > > -- > Sunil T T > Assistant Professor > Dept. of Electronics > College of Engineering > Attingal Pin 695 101 > http://brainstorms.in > > > > > --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~--- <>
[fsug-tvm] Re: Dual Cores
Hehehe, Please dont call me sir![?] On Sat, Mar 21, 2009 at 9:51 PM, Visakh wrote: > > > > On Mar 21, 7:38 pm, Syam wrote: > > Just another (probably offtopic) piece of informtion... The O(1) > scheduler > > is actually old. As of 2.6.23, the Completely Fair Scheduler (CFS) is > used > > in place of the O(1) scheduler. > > The scheduling complexity of CFS is O(log N). > > Thanks for pointing that out ശ്യാം ചേട്ടാ! I picked up the info from > an old IBM site and forgot the fact that CFS replaced O(1) recently. > But I don't think that it is an off-topic info. The choice of the > scheduler directly affects the answer to this topic. > > I found this simple clear explanation of scheduling in general, and O > (1) & CFS in particular: > http://immike.net/blog/2007/08/01/what-is-the-completely-fair-scheduler/ > . Though I can't find any comparison between O(1) and CFS on SMP > performance, the sheer simplicity and scalability of the newer CFS > compared to O(1) tells me that SMP support may be better. Infact some > reviews support this (3D games). So that's good news for Binny sir! > CFS uses an imaginary processor to decide on scheduling, and I see no > reason why it can't 'imagine up multiple processors'! :D. Of course, I > could be wrong- I am no kernel hacker! > > Regards, > Gokul Das > > > --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~--- <>
[fsug-tvm] Re: Dual Cores
Oops. My reply caused a lot of confusion. I was referring to kernel compilation as a possible use of multi core for a power user. See this https://help.ubuntu.com/community/Kernel/Compile Sunil On Sat, Mar 21, 2009 at 2:46 PM, Sunil Thomas Thonikuzhiyil < vu2...@gmail.com> wrote: > Depending on what you do. If you compile kernel you can speed up . > > Sunil > > > On Sat, Mar 21, 2009 at 1:52 PM, BINNY THOMAS wrote: > >> I want to know about the multiprocessor support of linux i.e. Dual >> cores,Core 2 Duo,Quad cores etc. Can it effectively utilise all the cores? >> >> Thanks >> >> >> >> > > > -- > Sunil T T > Assistant Professor > Dept. of Electronics > College of Engineering > Attingal Pin 695 101 > http://brainstorms.in > > -- Sunil T T Assistant Professor Dept. of Electronics College of Engineering Attingal Pin 695 101 http://brainstorms.in --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
On Mar 21, 7:38 pm, Syam wrote: > Just another (probably offtopic) piece of informtion... The O(1) scheduler > is actually old. As of 2.6.23, the Completely Fair Scheduler (CFS) is used > in place of the O(1) scheduler. > The scheduling complexity of CFS is O(log N). Thanks for pointing that out ശ്യാം ചേട്ടാ! I picked up the info from an old IBM site and forgot the fact that CFS replaced O(1) recently. But I don't think that it is an off-topic info. The choice of the scheduler directly affects the answer to this topic. I found this simple clear explanation of scheduling in general, and O (1) & CFS in particular: http://immike.net/blog/2007/08/01/what-is-the-completely-fair-scheduler/ . Though I can't find any comparison between O(1) and CFS on SMP performance, the sheer simplicity and scalability of the newer CFS compared to O(1) tells me that SMP support may be better. Infact some reviews support this (3D games). So that's good news for Binny sir! CFS uses an imaginary processor to decide on scheduling, and I see no reason why it can't 'imagine up multiple processors'! :D. Of course, I could be wrong- I am no kernel hacker! Regards, Gokul Das --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
On Sat, Mar 21, 2009 at 7:15 PM, Visakh wrote: > Linux SMP support is superb under 2.6 kernel series > after introduction of the new O(1) scheduler. > Just another (probably offtopic) piece of informtion... The O(1) scheduler is actually old. As of 2.6.23, the Completely Fair Scheduler (CFS) is used in place of the O(1) scheduler. The scheduling complexity of CFS is O(log N). -- Regards, Syam sya...@gmail.com --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Thanks for the information. Hope they start writing programs for multiprocesors as they are becoming common On Sat, Mar 21, 2009 at 7:15 PM, Visakh wrote: > > Hi, > > Syam Krishnan wrote: > > Why does one need to compile the kernel for effective multiprocessor > support?? > > The fact is, you dont really need to recompile the kernel in modern > Linux systems to get multicore support. The multicore I am referring > to is Symmetric Multi-Processor system (SMP) which includes the common > Intel Pentium D, Intel Core2 series and AMD X2 series. I am not sure > about other types of multicore processors like AMD Opteron (NUMA) and > IBM Cell (ASMP). Linux SMP support is superb under 2.6 kernel series > after introduction of the new O(1) scheduler. > > There are several ways to check if your kernel supports SMP: > 1) Simplest method: Check 'Resources' tab in the application 'System > Monitor' (System>Administration). It will show all the processors in > use and their loads. > > 2) Open terminal and try the command "grep -c ^processor /proc/ > cpuinfo" . If SMP enabled, you will get the number of processors in > use. > > 3) Open terminal and try "uname --all" . Most distributions will > indicate if they are using SMP-enabled kernel (Ubuntu does for me) > > And the answer for "Can it effectively utilise all the cores?". On any > OS (not specifically Linux), you can't take full advantage of SMP > processing power due to following reasons: > > i) Just the OS is not enough- the application has to be built > optimally for multithreading. Most applications we have today are not > like that. Maximum SMP efficiency is achieved only by the so called > 'embarrassingly parallel' applications (eg: compiling using distcc, > distributed ray tracing etc). That alone is one good reason why SMP > efficiency is low. > > ii) Our PCs and Laptops waste a lot of time I/O bound, waiting for > hard disk reads or user input. Actually this affect all PCs, not just > SMPs. But still, that work could have been done by a single processor. > > iii) There are overheads for Multiprocessing. More so, if the > applications are not built with SMPs in mind. > > In general, having N cores/processors is definitely better than having > just one. But it is never N times as fast as one core/processor. > Legacy applications make this situation even worse. > > Regards, > Gokul Das > > > --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Hi, Syam Krishnan wrote: > Why does one need to compile the kernel for effective multiprocessor support?? The fact is, you dont really need to recompile the kernel in modern Linux systems to get multicore support. The multicore I am referring to is Symmetric Multi-Processor system (SMP) which includes the common Intel Pentium D, Intel Core2 series and AMD X2 series. I am not sure about other types of multicore processors like AMD Opteron (NUMA) and IBM Cell (ASMP). Linux SMP support is superb under 2.6 kernel series after introduction of the new O(1) scheduler. There are several ways to check if your kernel supports SMP: 1) Simplest method: Check 'Resources' tab in the application 'System Monitor' (System>Administration). It will show all the processors in use and their loads. 2) Open terminal and try the command "grep -c ^processor /proc/ cpuinfo" . If SMP enabled, you will get the number of processors in use. 3) Open terminal and try "uname --all" . Most distributions will indicate if they are using SMP-enabled kernel (Ubuntu does for me) And the answer for "Can it effectively utilise all the cores?". On any OS (not specifically Linux), you can't take full advantage of SMP processing power due to following reasons: i) Just the OS is not enough- the application has to be built optimally for multithreading. Most applications we have today are not like that. Maximum SMP efficiency is achieved only by the so called 'embarrassingly parallel' applications (eg: compiling using distcc, distributed ray tracing etc). That alone is one good reason why SMP efficiency is low. ii) Our PCs and Laptops waste a lot of time I/O bound, waiting for hard disk reads or user input. Actually this affect all PCs, not just SMPs. But still, that work could have been done by a single processor. iii) There are overheads for Multiprocessing. More so, if the applications are not built with SMPs in mind. In general, having N cores/processors is definitely better than having just one. But it is never N times as fast as one core/processor. Legacy applications make this situation even worse. Regards, Gokul Das --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Sunil Thomas Thonikuzhiyil wrote: Depending on what you do. If you compile kernel you can speed up . Why does one need to compile the kernel for effective multiprocessor support?? Syam --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
[fsug-tvm] Re: Dual Cores
Depending on what you do. If you compile kernel you can speed up . Sunil On Sat, Mar 21, 2009 at 1:52 PM, BINNY THOMAS wrote: > I want to know about the multiprocessor support of linux i.e. Dual > cores,Core 2 Duo,Quad cores etc. Can it effectively utilise all the cores? > > Thanks > > > > -- Sunil T T Assistant Professor Dept. of Electronics College of Engineering Attingal Pin 695 101 http://brainstorms.in --~--~-~--~~~---~--~~ "Freedom is the only law". "Freedom Unplugged" http://www.ilug-tvm.org You received this message because you are subscribed to the Google Groups "ilug-tvm" group. To post to this group, send email to ilug-tvm@googlegroups.com To unsubscribe from this group, send email to ilug-tvm-unsubscr...@googlegroups.com For details visit the website: www.ilug-tvm.org or the google group page: http://groups.google.com/group/ilug-tvm?hl=en -~--~~~~--~~--~--~---
