Re: [Haifux] No! No! Don't compile your kernel!
On Sunday 23 Jan 2011 23:06:40 Nadav Har'El wrote: On Sun, Jan 23, 2011, Eli Billauer wrote about [Haifux] No! No! Don't compile your kernel!: Yet another sign that Linux is turning into a don't-touch-me kind of system. How many times did they tell me I don't really want to compile my kernel? http://wiki.centos.org/HowTos/Custom_Kernel When was the last time you compiled gcc on your own? I compiled it shortly after I read this article - http://lwn.net/Articles/387122/ in May 2010, and wanted to see how the new flags affect the speed of Freecell Solver, and it indeed benefited from them. The process was not hard and here's my script for it: [code] #!/bin/bash ~/Download/unpack/prog/gcc/gcc-4.5.0/configure \ --prefix=$HOME/apps/prog/gcc-4.5.0 \ --enable-languages=c,c++ [/code] (You need to build it in a different directory than the source.). Previously, in April 2009, I built gcc-2.95.3 (which is very old) to see how much faster it *compiles*, and how much slower the resultant executables are: http://tech.groups.yahoo.com/group/fc-solve-discuss/message/940 When was the last time you compiled the X Window System? It's been a while since I compiled the entire X Window System since it became modular, but I've built the x11-server on occasions for Mandriva. Though I used the .src.rpm. For me, the answers to both questions is 1995. If you answered similarly (or even, never), why should the kernel be any different - i.e., why do you need to compile a kernel unless you're a kernel developer (and 99.9% of Linux users aren't)? Well, I've also built some kernels for ocassions. The vanilla 2.6.37 kernel I built seemed snappier than the shipped-in Mandriva kernel, and it Freecell Solver executed there at 72.7685720920563s instead of 73.6936609745026s (the fractions are what was reported by my script and copy pasted here - they are not very accurate.). Before Linux had modules, you often needed to recompile the kernel to add new hardware. With the advent of kernel modules (in Linux 1.2 in 1995...), this is no longer the case. So really, why *would* you want to recompile the kernel, unless you are a kernel developer, i.e., modifying the kernel? Possibly to gain some speed or responsiveness or reduce memory consumption. Also, there are source-based distributions (e.g: Gentoo) which compile the source on each upgrade. Regards, Shlomi Fish -- - Shlomi Fish http://www.shlomifish.org/ Why I Love Perl - http://shlom.in/joy-of-perl Chuck Norris can make the statement This statement is false a true one. Please reply to list if it's a mailing list post - http://shlom.in/reply . ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] new Technion course: Operating Systems Engineering
Shachar Shemesh shac...@shemesh.biz wrote: On the one hand, this is an extremely interesting course. On the other hand, it sounds like a HUGE amount of work for just 4 points. You know what they say, no pain, no gain. This will be the sort of course that puts a beard on your face. Is the platform Intel, or something saner (say, ARM, PPC, MC680x0 or, come to think of it, just about any other CPU)? The platform is good ol' x86 because (1) everyone has one and (2) whatever you learn in this course will be immediately applicable in the real world. Cheers, Muli ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] new Technion course: Operating Systems Engineering
On Mon, Jan 24, 2011 at 12:09:19PM +0200, Shachar Shemesh wrote: Don't get me wrong. Had I still been a student, I'd take that class in a blink of an eye. I'm just worried that students will take it without realizing just how much work it is. We intend to clarify up front that this is going to require serious work, and anyone who isn't up to it, is welcome to drop the class. Is the platform Intel, or something saner (say, ARM, PPC, MC680x0 or, come to think of it, just about any other CPU)? The platform is good ol' x86 because (1) everyone has one and (2) whatever you learn in this course will be immediately applicable in the real world. It also has a horrific assembly, as well as a too many layers MMU which revives that segmented addresses everyone have been working so hard to forget ever existed. Undoubtedly x86 won't win any points in an ISA beauty contest, but it does what it does well, segmentation can be useful even today (see for example the original Xen protection model on 32 bit or vx32), and its hierarchical page table model is the one the students are familiar with and is heavily explored in the research literature. ARM, on the other hand, is (1) reasonable and straight forward assembly, and even reasonable MMU (2) has readily available emulators for all platforms and (3) is also fairly immediately applicable in the real world, arguably more so than X86. It also lags quite a bit behind x86 in various interesting ways, if you come from the server space: virtualization support for example, or efficient page table design. Not to mention that ARM apparently is following in the x86 foot steps -- didn't they just introduce the horrors of PAE? Let's put it this way. The chances a CS graduate will be asked to write an ARM based BSP are much higher than the chances she'll be asked to write an X86 based one. Maybe. But the chances he'll end doing kernel hacking on x86 are higher, especially if he lands eventually in our research group. Cheers, Muli ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] No! No! Don't compile your kernel!
On Mon, Jan 24, 2011, Shlomi Fish wrote about Re: No! No! Don't compile your kernel!: built seemed snappier than the shipped-in Mandriva kernel, and it Freecell Solver executed there at 72.7685720920563s instead of 73.6936609745026s (the I hope that you agree with me that 99.9218485921% of the users wouldn't bother themselves with recompilation (or any other manual step for that matter) to make their games run 1.27127529900685765% faster ;-) -- Nadav Har'El| Monday, Jan 24 2011, 19 Shevat 5771 n...@math.technion.ac.il |- Phone +972-523-790466, ICQ 13349191 |My password is my dog's name. His name is http://nadav.harel.org.il |a#j!4@h, but I change it every month. ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] No! No! Don't compile your kernel!
Just a couple of nitpicks, in the hope they will prove useful. On 24/01/11 09:02, Shachar Raindel wrote: dpkg-buildpackage -rfakeroot At least on debian, -rfakeroot is assumed unless you want something else (sudo, plugfakeroot-ng/plug or whatever). And if I skip the tinker stage, build is 100% sure to succeed. Let's agree on higher than 98%, so that we do not introduce unnecessary and unjustified hubris. Shachar -- Shachar Shemesh Lingnu Open Source Consulting Ltd. http://www.lingnu.com ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
[Haifux] My own posts to the list
Hi all, For some reason, I am not getting my own posts to the list. I don't know how it is when opening a new thread (which this message is testing), but when replying to a thread, I'm not getting the mail back from the mailing list server. I tried logging in to the list self maintenance interface, and I am set to receive my own posts. Any ideas? Shachar -- Shachar Shemesh Lingnu Open Source Consulting Ltd. http://www.lingnu.com ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] My own posts to the list
On 24/01/11 12:39, Shachar Shemesh wrote: Hi all, For some reason, I am not getting my own posts to the list. I don't know how it is when opening a new thread (which this message is testing), but when replying to a thread, I'm not getting the mail back from the mailing list server. I tried logging in to the list self maintenance interface, and I am set to receive my own posts. Any ideas? Shachar Apparently, this was a DNS problem that actually prevented my own posts from ever reaching the mailing list. Solved now (obviously), so please ignore. Shachar -- Shachar Shemesh Lingnu Open Source Consulting Ltd. http://www.lingnu.com ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] My own posts to the list
Just out of curiosity, how does DNS prevent a message from going through? On Jan 24, 2011 3:58 PM, Shachar Shemesh shac...@shemesh.biz wrote: On 24/01/11 12:39, Shachar Shemesh wrote: Hi all, For some reason, I am not getting my own posts to the list. I don't know how it is when opening a new thread (which this message is testing), but when replying to a thread, I'm not getting the mail back from the mailing list server. I tried logging in to the list self maintenance interface, and I am set to receive my own posts. Any ideas? Shachar Apparently, this was a DNS problem that actually prevented my own posts from ever reaching the mailing list. Solved now (obviously), so please ignore. Shachar -- Shachar Shemesh Lingnu Open Source Consulting Ltd. http://www.lingnu.com ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] My own posts to the list
On 24/01/11 16:00, Tzafrir Rehan wrote: Just out of curiosity, how does DNS prevent a message from going through? The server I use for outbound SMTP used to be the slave for several Hamakor domains. The new admins failed to notify me that that is no longer the case. As a result, it contained old A and MX records for haifux.org, resulting in trying to send all of my emails to the wrong machine (receiving connection refused, and queuing them for a while). Shachar -- Shachar Shemesh Lingnu Open Source Consulting Ltd. http://www.lingnu.com ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] No! No! Don't compile your kernel!
On Monday 24 Jan 2011 12:55:04 Nadav Har'El wrote: On Mon, Jan 24, 2011, Shlomi Fish wrote about Re: No! No! Don't compile your kernel!: built seemed snappier than the shipped-in Mandriva kernel, and it Freecell Solver executed there at 72.7685720920563s instead of 73.6936609745026s (the I hope that you agree with me that 99.9218485921% of the users wouldn't bother themselves with recompilation (or any other manual step for that matter) to make their games run 1.27127529900685765% faster ;-) shlomif[fcs]:~$ perl -E 'say ((73.6936609745026-72.7685720920563)/72.7685720920563*100)' 1.27127529900685 Kudos for the exact calculation. And yes, I agree that most people won't bother with this, but still the system running the vanilla kernel otherwise felt faster and snappier, and some people may wish to go to the extra trouble for that. (Without caring too much about the 1.27% speed increase in Freecell Solver's speed.). I don't know how much, and I'm possibly quite above the average computer user, but this option should still be relevant. I agree with you that most people should not bother with this and just use the packages that their distribution provides. All that put aside, as a software developer, I don't find 1% speed increases negligible, because when put together, they add up to significant savings. See what I wrote about it here: http://en.wikibooks.org/wiki/Optimizing_Code_for_Speed (Search for «Small» (with the double-quotes, but without the angle quotes). Regards, Shlomi Fish -- Nadav Har'El| Monday, Jan 24 2011, 19 Shevat 5771 n...@math.technion.ac.il |- Phone +972-523-790466, ICQ 13349191 |My password is my dog's name. His name is http://nadav.harel.org.il |a#j!4@h, but I change it every month. I think this joke would be funnier if it said and I change it every month. instead of but I change it every month.. Regards, Shlomi Fish -- - Shlomi Fish http://www.shlomifish.org/ Best Introductory Programming Language - http://shlom.in/intro-lang Chuck Norris can make the statement This statement is false a true one. Please reply to list if it's a mailing list post - http://shlom.in/reply . ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] No! No! Don't compile your kernel!
Hi, To put it short: I never was much into the ideals of freedom. My preference of free software always was because I could alter it to meet my own needs. It was easy enough to do for real. And I had this feeling that the system was meant to be hacked. It belonged to me. And that's fading away, most likely because nobody really seems to care about this. Linux is becoming a piece of opaque spaghetti, but it's OK as long as yum this or apt-get that trades one bug for another. Spaghetti is not free software. Not in any sense. Shachar Raindel wrote: And for both issues, if you were a company and not a home user, the risks of replacing a major software component with a version that haven't been tested and properly integrated with the rest of the system would have been far worse than the risks of root on NFS through initrd and small part of my hardware isn't supported by an extremely outdated system. Ah, that's my point. I clearly remember playing around with my kernel, upgrading it with a gap of several years with a vanilla kernel, and guess what? I just booted and all was OK. Would you believe that? When did it become so scary make changes in the kernel? And even worse: Why do we accept this? What's the practical difference between Window's kernel and Linux' if you can't compile one, but don't dare to compile the other? There are many kind of companies. My experience with companies, in particular large ones, is that they put stability before anything. If you have 1000 running stations out there, the last thing to do is to just upgrade everything in one go. The worst nightmare is some rare bug which suddenly knocks your entire infrastructure down (Cellcom, anyone?). Justified or not, large companies tend to stick to what they got for as long as possible. I mean, a lot of end-user terminals still emulate DOS-like screens in a window (Israeli railways? Banks?), because they stick to the interface they had 20 years ago. When small glitches mean big money, the top priorities are stability and control. Stability means to patch only what you need changed. Control means that you know what you're doing, to the smallest detail. yum upgrade and its apt parallel is losing control completely. Have you tried Ubuntu recently? If I have package X, and I want to tinker with its source files, all I have to do is: (...) Redhat has a similar system. I'm aware of it. It's yet another expression of the don't-touch-me paradigm. If you have to make a change in the sources, we'll hold your hands while you're doing it. The build process can be a piece of spaghetti, you won't understand it, it will work only on this specific distribution. And of course, if something fails, you're cooked. But why would it fail? Software never fails. It always lives up to these don't-worry promises that everything will be smooth, and you'll never need to understand how things work behind the scenes, and therefore it can be as messy as ever. How do you find information on the web? Who makes your Shampoo? Who is providing your Internet connectivity? We are dependent on big corporations for our everyday life whether you like it or not. You mean, like Microsoft? It looks like I'm the only one who want to throw away all those bells and whistles and go back to the good old days when you could do violent things to your OS and it would still run as if nothing happened. Or put simply: When GNU/Linux was stable as a rock. Eli -- Web: http://www.billauer.co.il ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] No! No! Don't compile your kernel!
On Monday, 24 בJanuary 2011 22:11:49 Eli Billauer wrote: ... It was easy enough to do for real. And I had this feeling that the system was meant to be hacked. It belonged to me. And that's fading away, most likely because nobody really seems to care about this. Linux is becoming a piece of opaque spaghetti, but it's OK as long as yum this or apt-get that trades one bug for another. Spaghetti is not free software. Not in any sense. 1. You should not be too worried about the lost art of kernel compilation from sources. Yes, only a tiny fraction of Linux users today compile their kernel (gcc, whatever)., but comparing fractions is a mistake. The *number* of people doing these compiles today in comparison with, say, 15 years ago is many-fold. It's simply that we now have many more people who are *only users* and should take their needs in account *as well* -- It doesn't mean we have less *developers* or that very few can do this the old way 2. The system in general *is* more complex today than 15 years ago. But attributing this purely to developers surrendering fashion is ignoring the real changes that affected Linux during this time: a lot more architectures, more cpus (numa), dynamic peripherals (scsi, usb, hot-plug pci, etc), hot-plugable cpus and ram (balloning) virtualization, embedded systems. And please note I only mentioned hardware related changes, ignoring functional changes (e.g: desktop integration) So not only you can compile your own stuff today, many of us do this pretty routinely (you cannot evade it completely in embdeded space yet). However, if you want to compile key parts of a modern *desktop* you'll simply have to work harder. BTW: the apt-get/yum mentioned before in this thread would also help you compile on your own because they can both bring you the build dependencies (and document them for you) and also contain the steps required for the build (which you can compare with your manual process if you have problems). Don't worry ;-) -- Oron Peled Voice: +972-4-8228492 o...@actcom.co.il http://users.actcom.co.il/~oron If you have an apple and I have an apple and we exchange apples then you and I will still each have one apple. But if you have an idea and I have an idea and we exchange these ideas, then each of us will have two ideas. -- George Bernard Shaw ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux
Re: [Haifux] No! No! Don't compile your kernel!
(top-posting. yey!) just as an example - we are using CentOS at work. We needed to touch something in the kernel - so we changed the code and compiled it. we Stumbled into bugs in various networking-related packages - so we took the srpms, read the source and fixed some bugs. we had a kernel crash bug - and we had a dump that allowed us to find the cause of the bug - and sending a report to the relevant mailing list - someone made a fix, and we could back-port it into our (centos) kernel and solve the issue. we had varius problems with python, so we could delve into python's source code and understand what is causing them. we also read parts of the code of glibc to understand why some APIs don't work as we expect them to work. so, linux is still open source, when you're on the side of the developers. it takes more then in the past to do things, because the system is more complicated and contains much more packages and much more lines of code. --guy Original message Date: Mon, 24 Jan 2011 23:12:48 +0200 From: Oron Peled o...@actcom.co.il Subject: Re: [Haifux] No! No! Don't compile your kernel! To: haifux@haifux.org Cc: Shachar Raindel shach...@gmail.com On Monday, 24 בJanuary 2011 22:11:49 Eli Billauer wrote: ... It was easy enough to do for real. And I had this feeling that the system was meant to be hacked. It belonged to me. And that's fading away, most likely because nobody really seems to care about this. Linux is becoming a piece of opaque spaghetti, but it's OK as long as yum this or apt-get that trades one bug for another. Spaghetti is not free software. Not in any sense. 1. You should not be too worried about the lost art of kernel compilation from sources. Yes, only a tiny fraction of Linux users today compile their kernel (gcc, whatever)., but comparing fractions is a mistake. The *number* of people doing these compiles today in comparison with, say, 15 years ago is many-fold. It's simply that we now have many more people who are *only users* and should take their needs in account *as well* -- It doesn't mean we have less *developers* or that very few can do this the old way 2. The system in general *is* more complex today than 15 years ago. But attributing this purely to developers surrendering fashion is ignoring the real changes that affected Linux during this time: a lot more architectures, more cpus (numa), dynamic peripherals (scsi, usb, hot-plug pci, etc), hot-plugable cpus and ram (balloning) virtualization, embedded systems. And please note I only mentioned hardware related changes, ignoring functional changes (e.g: desktop integration) So not only you can compile your own stuff today, many of us do this pretty routinely (you cannot evade it completely in embdeded space yet). However, if you want to compile key parts of a modern *desktop* you'll simply have to work harder. BTW: the apt-get/yum mentioned before in this thread would also help you compile on your own because they can both bring you the build dependencies (and document them for you) and also contain the steps required for the build (which you can compare with your manual process if you have problems). Don't worry ;-) -- Oron Peled Voice: +972-4-8228492 o...@actcom.co.il http://users.actcom.co.il/~oron If you have an apple and I have an apple and we exchange apples then you and I will still each have one apple. But if you have an idea and I have an idea and we exchange these ideas, then each of us will have two ideas. -- George Bernard Shaw ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux ___ Haifux mailing list Haifux@haifux.org http://hamakor.org.il/cgi-bin/mailman/listinfo/haifux