Thread: [SSI-users] Binary release and future kernel development
Brought to you by:
brucewalker,
rogertsang
|
From: Brian J. W. <Bri...@hp...> - 2003-04-03 00:08:59
|
The 0.9.6 binary release should be ready tomorrow. It seems run well on my test cluster, but there's a bug John would like to fix before releasing it. Some of you might wonder why it takes so long between doing the source release and doing the binary release. The reason is that the source release is based on plain vanilla Linux 2.4.18, whereas the binary release is based on Red Hat's 2.4.18-18 kernel. The two have significant differences that affect the OpenSSI implementation. The latest feature enhancements and bug fixes from the vanilla-based version of OpenSSI do not always merge cleanly with the Red Hat-based version, and the conflicts are typically in code that someone else developed. I have to make intelligent guesses and consult with the appropriate developer about how to resolve each conflict. Even after this, there are still compile-time and run-time bugs to be fixed. As a result I suspect the binary release is less stable than the source release. So now you might be wondering why the source and binary releases use different code bases. The answer is that in the beginning we were doing just source releases and wanted to be distribution-agnostic by basing it on the vanilla kernel. This was fine for development, but not necessarily for an end-user. A few months ago when Bruce asked me to roll our first binary release for end-users, he realized they would not want to give up the bug fixes and additional features that Red Hat has and vanilla Linux doesn't. He decided the binary release should be based on the latest Red Hat kernel. Lately I've been thinking it would be best to do all development on the Red Hat kernel to avoid the merging and instability problems described above. I'm not suggesting that we abandon the vanilla kernel; if someone wants to take the responsibility to periodically merge new code with it and roll releases based on it, I'd be more than happy to tell you how to do that. Let me know what you think, Brian |
|
From: Aneesh K. K.V <ane...@di...> - 2003-04-03 03:16:07
|
Hi Brian, Firstly by doing this we are going to have problems with architectures that redhat doesn't support. I am worried about alpha here. As of now their kernel build and runs on alpha. Thanks to some people still interested in alpha inside redhat. But i am not sure for how long they are going to do that. Another important thing to note is that by making the development against redhat kernel we are making it more difficult to port open-ssi to other architectures. ( One of my friend was talking to me about sparc... ) Redhat add many bug fixes and also do back porting of many new features into the kernel. But i don't think any bug fix which the user really want is missed from the vanilla kernel. Regarding enhancement the entire kernel team consider most of these changes not ready for 2.4. If it is difficult to get the changes merged back to redhat kernel, I would say we can make binary releases of vanilla kernel rather than switching to redhat kernels. I can say by looking at open-mosix download statistics, lot of people will still be using the binary release of vanilla kernel. If you really see the early users of open-ssi for that matter any project than brings changes to kernel, they all are people who can afford more than one machines for trying out new stuff. This include people who are rich enough to afford many machine ( I don't have even one :( ) , students and research people. I don't think any of this class worry about redhat patches. Do we have anybody running open-ssi on production systems ?. Following the vanilla kernel make sure at some point in future there is a chance of getting merged with main line Linux kernel. -aneesh On Thu, 2003-04-03 at 05:40, Brian J. Watson wrote: > The 0.9.6 binary release should be ready tomorrow. It seems run well on > my test cluster, but there's a bug John would like to fix before > releasing it. > Some of you might wonder why it takes so long between doing the source > release and doing the binary release. The reason is that the source > release is based on plain vanilla Linux 2.4.18, whereas the binary > release is based on Red Hat's 2.4.18-18 kernel. The two have significant > differences that affect the OpenSSI implementation. The latest feature > enhancements and bug fixes from the vanilla-based version of OpenSSI do > not always merge cleanly with the Red Hat-based version, and the > conflicts are typically in code that someone else developed. I have to > make intelligent guesses and consult with the appropriate developer > about how to resolve each conflict. Even after this, there are still > compile-time and run-time bugs to be fixed. As a result I suspect the > binary release is less stable than the source release. > > So now you might be wondering why the source and binary releases use > different code bases. The answer is that in the beginning we were doing > just source releases and wanted to be distribution-agnostic by basing it > on the vanilla kernel. This was fine for development, but not > necessarily for an end-user. A few months ago when Bruce asked me to > roll our first binary release for end-users, he realized they would not > want to give up the bug fixes and additional features that Red Hat has > and vanilla Linux doesn't. He decided the binary release should be based > on the latest Red Hat kernel. > > Lately I've been thinking it would be best to do all development on the > Red Hat kernel to avoid the merging and instability problems described > above. I'm not suggesting that we abandon the vanilla kernel; if someone > wants to take the responsibility to periodically merge new code with it > and roll releases based on it, I'd be more than happy to tell you how to > do that. > > Let me know what you think, > > Brian |
|
From: Brian J. W. <Bri...@hp...> - 2003-04-05 03:16:58
|
Aneesh Kumar K.V wrote: >Do we have anybody running open-ssi on production systems ?. > Perhaps no one's yet using it in production, but I suspect there are a few people silently watching these lists and evaluating how they could do so. Most of them probably want to run Red Hat. For their sake, it's important to make sure there are stable OpenSSI RPMs that are as close as possible to their base distribution. That's why I want to focus my efforts on the Red Hat branch, to make sure that it's stable and receives sufficient testing. As I said, I'm not suggesting we abandon the vanilla kernel. I'm confident someone will step up to make sure the trunk of the repository stays in sync with the Red Hat branch. If anyone chooses to continue development on the trunk, I will similarly make sure that important changes are pulled over to the RH side. As you say, development of Alpha is a good thing to continue on the trunk, since RH may not continue to support it. For the same reason, I won't worry about pulling Alpha stuff over to the RH branch. On the other hand, LVS is something I do want keep synced, and I will make sure that happens if you want to develop it on the trunk. >Following the vanilla kernel make sure at some point in future there is >a chance of getting merged with main line Linux kernel. > > This is more of 2.5 thing, which will require a merge anyway. The merges from 2.4 to RH were not horribly difficult, but they did require some effort and testing. Getting into the main line kernel is a bridge we can cross in the future. Brian |
|
From: Aneesh K. K.V <ane...@di...> - 2003-04-05 11:59:42
|
On Sat, 2003-04-05 at 08:48, Brian J. Watson wrote: > Aneesh Kumar K.V wrote: > > >Do we have anybody running open-ssi on production systems ?. > > > > Perhaps no one's yet using it in production, but I suspect there are a > few people silently watching these lists and evaluating how they could > do so. Most of them probably want to run Red Hat. For their sake, it's > important to make sure there are stable OpenSSI RPMs that are as close > as possible to their base distribution. That's why I want to focus my > efforts on the Red Hat branch, to make sure that it's stable and > receives sufficient testing. We didn't hear from any of them. But there were two people who requested not to switch to Redhat kernel. So the voting is 2:0 :) > > As I said, I'm not suggesting we abandon the vanilla kernel. I'm > confident someone will step up to make sure the trunk of the repository > stays in sync with the Red Hat branch. If anyone chooses to continue > development on the trunk, I will similarly make sure that important > changes are pulled over to the RH side. > > As you say, development of Alpha is a good thing to continue on the > trunk, since RH may not continue to support it. For the same reason, I > won't worry about pulling Alpha stuff over to the RH branch. On the > other hand, LVS is something I do want keep synced, and I will make sure > that happens if you want to develop it on the trunk. > That brings other problem. John has been really good at taking care of all architecture when moving to higher kernel releases. Whenever he does that he takes care of all the archs. ( Great work!!!). Infact i got 2.4.18 up on alpha with just one compile fix. Now if you don't pull alpha to RH branch that means John/others will not be doing the merge when we go up the releases and alpha will need more work than now. This is going to be true for all other architecture that RH doesn't support. If the interest in IA64 decrease possibly that too ;-) So what we achieve by this is, to get those big boys who hesitate to run vanilla kernel on their box try and use open-ssi we are making porting/supporting of open-ssi on other architectures more difficult. Again from what others/projects are doing i see most of them working on vanilla linux kernel . But One thing I would like to say is that, if you are sure that we can do early/frequent releases and easy installable STABLE RPMS by switching the redhat kernel, then my vote is to switch to redhat kernel. That make the voting result 2:1 ;-) -aneesh |
|
From: Brian J. W. <Bri...@hp...> - 2003-04-15 02:24:42
|
Aneesh Kumar K.V wrote: > We didn't hear from any of them. But there were two people who requested > not to switch to Redhat kernel. So the voting is 2:0 :) I'm not suggesting the entire OpenSSI project switch to the Red Hat kernel. My concern is that I don't have enough time to support both the vanilla and Red Hat kernels. I'm more interested in the Red Hat kernel, so I'm choosing to support that. The OpenSSI project is a community effort. If the community is interested in supporting the vanilla kernel, then someone from the community will step up to make sure it happens. :) I can help him/her by describing the procedures I use to roll releases and merge changes from one branch to another. After he demonstrates the quality of his work, I can give him full admin access so he can support the vanilla kernel without any help from me. Doing this is a good way for someone new to contribute to the project, and it frees up my time to pursue what I want to do. ;) > That brings other problem. John has been really good at taking care of > all architecture when moving to higher kernel releases. Whenever he does > that he takes care of all the archs. ( Great work!!!). If John doesn't merge with new vanilla kernels anymore, that just means someone else will have to do it. :) It's not necessarily that difficult. CVS does most of the work. For any difficult conflicts that arise during the merge, I'm sure John, myself, or any other developer can offer advice on how to resolve the conflict. > Again from what others/projects are doing i see most of them working on > vanilla linux kernel . Most kernel projects are drivers and filesystems. I doubt they have serious portability problems between vanilla and distribution-specific kernels. Of those that are not drivers or filesystems, most only touch one or two subsystems. Therefore it's not terribly difficult for a talented user of an average kernel project to apply their patch against a Red Hat kernel and get it to work. OTOH, OpenSSI has to touch almost everything to create the illusion that the cluster is a single machine. Without some knowledge about how OpenSSI works, it's virtually impossible for the same user to accomplish the same feat with its vanilla kernel patch. I know this because even John had to take a few days to do the initial port to the RH kernel. ;) > But One thing I would like to say is that, if you are sure that we can > do early/frequent releases and easy installable STABLE RPMS by switching > the redhat kernel, then my vote is to switch to redhat kernel. That make > the voting result 2:1 ;-) The RPM releases will definitely be much easier to do, which will allow me to do them more often. Hopefully they will also be more stable. The more development that's done on the RH branch, the more likely this will be true. ;) Brian |
|
From: Eduardo K. <ed...@li...> - 2003-04-05 06:10:00
|
On Fri, Apr 04, 2003 at 07:18:11PM -0800, Brian J. Watson wrote: > > Perhaps no one's yet using it in production, but I suspect there are a > few people silently watching these lists and evaluating how they could > do so. Most of them probably want to run Red Hat. For their sake, it's > important to make sure there are stable OpenSSI RPMs that are as close > as possible to their base distribution. That's why I want to focus my > efforts on the Red Hat branch, to make sure that it's stable and > receives sufficient testing. > I second that. I am doing just what you describe :) reading and waiting... -- Eduardo Kaftanski ed...@li... Gerente Ingenieria LinuxCenter S.A. Canada 239 5to Piso, Providencia, Stgo de Chile. http://www.linuxcenter.cl 2745000 |
|
From: Christian L. <ly...@po...> - 2003-04-09 17:05:47
|
Ops... perhaps I'm late, but my system is only waiting users, so it's almost in production!!! BTW, I'm using Debian, so I'm against to switch to a redhat kernel (now it's 3:0) :-) On Saturday 05 April 2003 03:09, Eduardo Kaftanski wrote: > On Fri, Apr 04, 2003 at 07:18:11PM -0800, Brian J. Watson wrote: > > Perhaps no one's yet using it in production, but I suspect there are a > > few people silently watching these lists and evaluating how they could > > do so. Most of them probably want to run Red Hat. For their sake, it's > > important to make sure there are stable OpenSSI RPMs that are as close > > as possible to their base distribution. That's why I want to focus my > > efforts on the Red Hat branch, to make sure that it's stable and > > receives sufficient testing. > > I second that. I am doing just what you describe :) > reading and waiting... -- Christian Lyra POP-PR - RNP http://lyra.soueu.com.br The Tao Of Programing |
|
From: Brian J. W. <Bri...@hp...> - 2003-04-15 01:51:56
|
Christian Lyra wrote: > Ops... perhaps I'm late, but my system is only waiting users, so it's almost > in production!!! Excellent!! Let us know how it works out, if you're at liberty to disclose what you're doing with it. ;) Brian |
|
From: Gustavo <gu...@in...> - 2004-01-13 13:50:24
|
Eduardo Kaftanski wrote: > On Fri, Apr 04, 2003 at 07:18:11PM -0800, Brian J. Watson wrote: > >>Perhaps no one's yet using it in production, but I suspect there are a >>few people silently watching these lists and evaluating how they could >>do so. Most of them probably want to run Red Hat. For their sake, it's >>important to make sure there are stable OpenSSI RPMs that are as close >>as possible to their base distribution. That's why I want to focus my >>efforts on the Red Hat branch, to make sure that it's stable and >>receives sufficient testing. >> > > > I second that. I am doing just what you describe :) > reading and waiting... > > Also, testing, reading, and waiting! Gustavo. |