Announcement

Collapse
No announcement yet.

Linux hacker compares Solaris kernel code:

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #46
    Originally posted by i386reaper View Post
    I can't agree more. In fact there are a lot of people who agree with Pawlerson the BSD and Solaris = Shit:

    http://aboutthebsds.wordpress.com/20.../bsd-vs-linux/

    BTW, this blog is really interesting and enlightening. I'd recommend it to everyone, especially those pro-BSD fools.
    Quite funny to read, but there's a lot of truth in this article. However, I'd prefer it to be more serious, because some parts don't reflect the reality.

    Comment


    • #47
      Originally posted by Pawlerson View Post
      Yeah, right. HTC people care about OS price. It also a fact Solaris has higher memory footprint and introduces higher overhead in comparison to Linux. You don't have to believe me, but you can check this yourself and you can even find about this in google.
      This is true, but it's also true of mainstream Linux. This is why IBM and Cray have their own cut down operating systems that are run on the compute nodes of large scale HPC systems (AFAIK, the IBM one is not based on Linux, but the Cray one is).

      The front end nodes often run Linux, but I believe in the case of some IBM systems, AIX is used instead. Probably depends on what the customer wants.

      Comment


      • #48
        Originally posted by Pawlerson View Post
        I'm not blaming him, because of BFS or his work. I don't like his attitude.
        You usually need to hear the accused's side of the story too before forming a final opinion:

        http://apcmag.com/interview_with_con...he_desktop.htm

        Comment


        • #49
          Originally posted by ChrisXY View Post
          So this is another point I don't get. Linux doesn't scale to 8 CPUs yet the supercomputer at my university with more than 100.000 cores (I believe 2 16 core CPUs per node) runs Linux. Sure, there may very well be other systems that scale better, but saying linux has a "hard time"? Come on!
          You must differentiate between different types of scaling. Linux scales very well on clusters, large networks with embarassingly parallell workloads, just like supercomputers. For isntance, the SGI Altix Linux server with 2048 cores, is a cluster. These are called HPC servers (i.e. cluster). And these supercomputers do not run stock Linux, they run heavily modified Linux. For instance the IBM Blue Gene supercomputer, runs Linux to distribute the work load out to each node, and then each node use another OS to do the simple computing. No complex things going on here, just number crunching.

          On the other hand, Linux scales very bad on SMP servers. That is, one big fat server with as many as 32 or even 64 cpus, for instance, IBM mainframes, Oracle M9000/M5, HP Integrity/Superdome. There are no Linux SMP servers for sale today, with more than 8 cpus. Linux has a very hard time scaling to 8 cpus. These big SMP servers, typically weigh 1000kg or more, and costs many millions. For instance, the huge IBM P595 server, with 32 cpus, used for the old TPC-C benchmark, costed $35 million list price. Not a typo. One single server costed $35 million.

          So I say like I said in the other thread: show me a Linux SMP server for sale with more than 8 cpus. Last I checked, there was no vendor that sold large Linux SMP servers. The question is: why? These SMP servers costs millions and if Linux can offer a 32 cpu server for a fraction of the price, then everybody would buy. Clearly there is a market opportunity, but no Linux vendor is offering SMP servers. Why? Is it because no one wants to earn millions, or is it because Linux does not scale? Hint: see 8-socket benchmarks with Linux and discover yourself how bad results Linux gets.

          Comment


          • #50
            Don't forget about who you're talking to.

            http://phoronix.com/forums/showthrea...767#post259767

            Comment


            • #51
              So what? I was alone in that thread and there was lot of people throwing all questions at me. I simply did not have time to continue. I mean, when you have been "debating" with Kraftman (Pawlerson, etc) and the like - you will see that it is pointless to continue. Even when you post research papers, they will not be convinced. Have you tried to talk to lot of people at the same time, including people like Kraftman? Do you know how many hours it takes? When I write a long post, it takes me 2-3 hours, because I google links and white papers, etc. I do research. I do not post unverifiable claims. When I post 10 long posts, it takes easily 2 days. I dont have that time to do that.

              Regarding that post you link to: yes, it was long. But you know what? He never answered to my question: why are there no 16 cpu Linux servers for sale? There are only 8 socket Linux servers for sale, and 2048 cpu clusters (SGI Altix and the like), but there are no 16 or 32 cpu socket Linux for sale. If you read his post, he never answered to that question, he ducked that question every time I asked it. Why? So, can somebody show me a 16 or 32 Linux SMP server for sale today? Or are you going to duck this question again?

              And he claimed lot of things, that there are no SMP servers today. How does he know that? For instance, this server is SMP
              http://www.oracle.com/us/products/se...iew/index.html
              but he says: "no that is not true, it is not a SMP server, it is a NUMA server". Why should I believe him? Have he really checked that all servers are NUMA today? For instance, the old IBM Mainframes that are sold even today, are SMP servers. "There are no SMP servers today, all are NUMA" - yeah right. How can he prove that? If I would post such a claim, I would have posted links that shown that each big 32 cpu server is NUMA - then I would be able to conclude "there are no SMP servers today" - but he never did that. Instead he claimed false things, and ducked my question.

              And again: where are those 16 or 32 cpu Linux servers for sale? How many times can people here duck that question? Actually, I have asked this question maybe... 50? times here in different threads, and I have never. ever. got. an. answer. I think it is funny. Benchmarks show Linux have performance problems on 8-socket servers, and there are no 16 or 32 cpu linux servers for sale. And no one have discussed this, everybody have ducked this.

              For the 51th(?) time: where are those 16 cpu Linux servers for sale? Where are they? Can someone answer me this? Have you tried to ask a question 50 times, to different persons, and never got an answer? Wouldnt you wonder why? Where are those servers? Where? Someone? It just takes some links, and I will stop asking this question. But no one have posted links. I will ask this question 50 times more, but no one will ever show me 16 cpu Linux servers. Why? Because they dont exist. Why? Because Linux can not scale to 16 cpus.

              Comment


              • #52
                Originally posted by kebabbert View Post
                So what? I was alone in that thread and there was lot of people throwing all questions at me. I simply did not have time to continue. I mean, when you have been "debating" with Kraftman (Pawlerson, etc) and the like - you will see that it is pointless to continue. Even when you post research papers, they will not be convinced. Have you tried to talk to lot of people at the same time, including people like Kraftman? Do you know how many hours it takes? When I write a long post, it takes me 2-3 hours, because I google links and white papers, etc. I do research. I do not post unverifiable claims. When I post 10 long posts, it takes easily 2 days. I dont have that time to do that.

                Regarding that post you link to: yes, it was long. But you know what? He never answered to my question: why are there no 16 cpu Linux servers for sale? There are only 8 socket Linux servers for sale, and 2048 cpu clusters (SGI Altix and the like), but there are no 16 or 32 cpu socket Linux for sale. If you read his post, he never answered to that question, he ducked that question every time I asked it. Why? So, can somebody show me a 16 or 32 Linux SMP server for sale today? Or are you going to duck this question again?

                And he claimed lot of things, that there are no SMP servers today. How does he know that? For instance, this server is SMP
                http://www.oracle.com/us/products/se...iew/index.html
                but he says: "no that is not true, it is not a SMP server, it is a NUMA server". Why should I believe him? Have he really checked that all servers are NUMA today? For instance, the old IBM Mainframes that are sold even today, are SMP servers. "There are no SMP servers today, all are NUMA" - yeah right. How can he prove that? If I would post such a claim, I would have posted links that shown that each big 32 cpu server is NUMA - then I would be able to conclude "there are no SMP servers today" - but he never did that. Instead he claimed false things, and ducked my question.

                And again: where are those 16 or 32 cpu Linux servers for sale? How many times can people here duck that question? Actually, I have asked this question maybe... 50? times here in different threads, and I have never. ever. got. an. answer. I think it is funny. Benchmarks show Linux have performance problems on 8-socket servers, and there are no 16 or 32 cpu linux servers for sale. And no one have discussed this, everybody have ducked this.

                For the 51th(?) time: where are those 16 cpu Linux servers for sale? Where are they? Can someone answer me this? Have you tried to ask a question 50 times, to different persons, and never got an answer? Wouldnt you wonder why? Where are those servers? Where? Someone? It just takes some links, and I will stop asking this question. But no one have posted links. I will ask this question 50 times more, but no one will ever show me 16 cpu Linux servers. Why? Because they dont exist. Why? Because Linux can not scale to 16 cpus.
                Well, for one there's that M5-32 with Oracle VM Server--which is a Linux-Xen system.
                But that's probably not what you meant.

                And that is NUMA; if you don't believe me, believe Oracle:
                Non-uniform memory access optimization in Oracle Solaris. With memory managed by each
                SPARC M5 processor, these implementations represent a non-uniform memory access (NUMA)
                architecture. In NUMA architectures, the time needed for a processor to access its own memory is
                slightly shorter than that required to access memory managed by another processor.
                (Quoting page 12 of http://www.oracle.com/technetwork/se...re-1920556.pdf)

                Comment


                • #53
                  Originally posted by Pawlerson View Post
                  Quite funny to read, but there's a lot of truth in this article. However, I'd prefer it to be more serious, because some parts don't reflect the reality.
                  What truth? The guy clearly has no technical knowledge if he couldn't get to a command line login in FreeBSD.

                  Comment


                  • #54
                    Originally posted by OpenSLOWlaris
                    In the 24 post in this thread: http://phoronix.com/forums/showthrea...nchmarks/page3

                    Kebabbart's true name is Orvar Korvar.



                    So, Kebabbart == Orvar Korvar. What a stupid name, fitting for an anti-Linux/pro-proprietary troll. Seriously, who gave you that name, the nurse at the orphanage in which your teenage mom dumped you because you were so ugly?
                    Hahahahahahaha. Shut up you fucking loser.

                    Comment


                    • #55
                      Originally posted by OpenSLOWlaris
                      In the 24 post in this thread: http://phoronix.com/forums/showthrea...nchmarks/page3

                      Kebabbart's true name is Orvar Korvar.

                      So, Kebabbart == Orvar Korvar. What a stupid name, fitting for an anti-Linux/pro-proprietary troll...
                      You probably know what I think about Linux vs Solaris (I'd enjoy a Pentium with Linux more than an Opteron with Solaris), but your spew still makes me sick.
                      Here's a hint: FUD is not long-term sustainable...see the "Halloween Documents" for an illustration.

                      Anyhow: I looked at the benchmarks mentioned there. I assume these are the basis of the "Linux starts to degrade at 8 cpus" claim.
                      It's two systems, Sunfire/Solaris 10 vs Proliant/SLES 10, and the basis of the claim appears to be that the Sunfire gets noticeably better response times and has higher CPU utilization, and is thus Solaris is better at scheduling.
                      Looking at the specs, it looks to me that it's actually a question of memory bound vs cpu bound: the Sunfire has 256 GB of RAM vs the Proliant's 128 GB.
                      Besides that, there are different RDBMSs in use (Oracle vs MaxDB), and the Sunfire has a slightly slower CPU.

                      And of course there's one more point:
                      Linux in 2009 is not the Linux of today. SLES 10 used kernel 2.6.16, and current is 3.9; there have been several changes in the scheduler since then.

                      Comment


                      • #56
                        Originally posted by OpenSLOWlaris
                        What a stupid name, fitting for an anti-Linux/pro-proprietary troll. Seriously, who gave you that name, the nurse at the orphanage in which your teenage mom dumped you because you were so ugly?
                        Did your mum drop you on your head?
                        Seriously, you don't agree that ogg should be permissively licensed because it might increase uptake, even though RMS and everyone else agrees with it.
                        So we've already worked out that you have absolutely no life, and you post links that don't back up your claims. In fact, the one you supplied didn't link to anything.
                        Do you program?
                        Do you even have proof that you program?
                        If you don't, what gives you the right to viciously criticise the works of others and not just to the creator?

                        Comment


                        • #57
                          Ok, I mistook 8 cpus for 8 cores. There are not many 8+ socket servers to find at all with google.

                          Here is one up to 32 socket one with HP-UX or linux: http://h18000.www1.hp.com/products/q.../11717_na.HTML

                          For benchmarks I only found the blog like website of this guy: http://www.c0t0d0s0.org/archives/619...P-and-Sun.html
                          But they never tested on the same machine so I wouldn't give too much. Also they are all 4-5 years old.

                          Any recent benchmarks on the same machine?

                          edit: IBM's System z series is also provided with Linux. I think it is mostly expensive for all the special hardware for I/O and the backwards compatible zOS and stuff like that.
                          Last edited by ChrisXY; 05-10-2013, 05:06 AM.

                          Comment


                          • #58
                            As I am also from sweden (as Kebabbert) I can tell you that no, his name is not 'Orvar Korvar', which is a (imo boring) 'joke' name in the same vein as Kebabbert (kebab + bert). As for his Solaris zelotry, as I recall he first found the light with Solaris when Sun was open sourcing their 'technologies'.

                            He (kebabbert) was (or atleast claimed to be) a proponent of open source and he thought this was a great move by Sun, but like is the case with certain fervent supporters (asperger syndrome?), they can totally identify themselves with something, which in this case is Solaris. Hence even now when Open Solaris is all but dead, and Oracle is putting their energy behind Linux, Kebabbert still is carrying on his fight against Linux, which he sees as Solaris 'enemy' (ironic in how Linux is also open source which is what once attracted him to Solaris).

                            So he is still searching the web for anything which he may spin as 'proof' of Solaris superiority, and sad as that may be it's certainly no worse than the pure trolling of however is behind the 'slowlaris','bsdsucks' etc accounts.

                            But just as trying to have a conversation with 'slowlaris' is totally pointless, so is trying to have a conversation with kebabbert, you can point to any number of facts, he can lose every argument (as in the linked previous discussion with 'TheOrqwithVagrant'), but he will just lick his wounds and be back. Logic or arguments will never sway him, you'd have a better shot at having RMS embrace proprietary software or Ballmer to open source Windows (under GPL even).

                            Comment


                            • #59
                              Thanks for trying to refute my point with facts.

                              Originally posted by ChrisXY View Post
                              Ok, I mistook 8 cpus for 8 cores. There are not many 8+ socket servers to find at all with google.

                              Here is one up to 32 socket one with HP-UX or linux: http://h18000.www1.hp.com/products/q.../11717_na.HTML
                              This server is a HP server made for HP-UX. They have compiled Linux to it, and offer Linux on it. But it is not a Linux server. It is a HP-UX server.


                              For benchmarks I only found the blog like website of this guy: http://www.c0t0d0s0.org/archives/619...P-and-Sun.html
                              But they never tested on the same machine so I wouldn't give too much. Also they are all 4-5 years old.
                              This benchmark shows that Linux used the same cpus, running at 2.8GHz. Solaris used the same cpus, running at 2.6GHz.


                              Any recent benchmarks on the same machine?
                              I showed a link with benchmarks on same hardware.


                              edit: IBM's System z series is also provided with Linux. I think it is mostly expensive for all the special hardware for I/O and the backwards compatible zOS and stuff like that.
                              z/Linux has been ported, but runs ontop z/OS (as I understand it) virtualized. Mainframes are not Linux servers, they are running z/OS.



                              See? My point is that there are no Linux servers offering 16 or 32 cpus for sale. Why? These servers costs many millions, and clearly there is a market opportunity. If RedHat or someone could sell such a Linux server for only half a million, all large investment banks, telcos, etc would switch at once. But no, there are no such servers. Why is that? You tell me. Nobody does not want to be rich?

                              Comment


                              • #60
                                Originally posted by Ibidem View Post
                                Looking at the specs, it looks to me that it's actually a question of memory bound vs cpu bound: the Sunfire has 256 GB of RAM vs the Proliant's 128 GB. Besides that, there are different RDBMSs in use (Oracle vs MaxDB), and the Sunfire has a slightly slower CPU.

                                And of course there's one more point:
                                Linux in 2009 is not the Linux of today. SLES 10 used kernel 2.6.16, and current is 3.9; there have been several changes in the scheduler since then.
                                The Linux 128 GB server used faster RAM dimms, that is the reason Linux used 128GB. You can install 256GB RAM on that server, but the benchmarking team chose not to.

                                And Solaris 11 has changed too compared to Soalris 10, just as Linux has.

                                Comment

                                Working...
                                X