| fernand0 | Hello, |
|---|---|
| fernand0 | our next speaker is David Santo Orcero. |
| fernand0 | He is an active developer and researcher who lives in Málaga, Spain. |
| fernand0 | He is a well known writer for several linux-related (and other) |
| fernand0 | journals. |
| fernand0 | He is one of the pioneers in this conference since he has participated in |
| fernand0 | all the three editions, with very interesting talks. |
| fernand0 | He is actively involved in the OpenMosix project and he is going to talk about "The SSI clustering proyect Open Mosix" |
| fernand0 | Thank you to him for comming here and preparing his talk for today. Thank |
| fernand0 | you to all for comming here. |
| fernand0 | We need some volunteers for Spanish->English translation |
| fernand0 | If you feel you can help with this, please contact mjesus |
| fernand0 | As usual, questions and comments can be done in #qc |
| fernand0 | David... |
| irbis | Hello, all! |
| irbis | I am going to give you a talk about the SSI clustering project OpenMosix. |
| irbis | First of all, I must introduce myself. |
| irbis | My name is David Santo Orcero. I am the maintainer of the userland part of OpenMosix project. |
| irbis | Some of you maybe know me because of my articles, or because my other speaks about clustering stuff. |
| irbis | I began to became interested on clustering on 1996, when I began to work on my Ms.Sc. |
| irbis | Since then, I have been learning about clustering. |
| irbis | Then, let's go. |
| irbis | You are probably more interested on SSI rather than on this introductory stuff. ;-) |
| irbis | On my talk, I am not to use the world "free", because is potentially dangerous. |
| irbis | Some people, including most zealots, slashdotters and other Yhad guys think on "free beer". |
| irbis | I will use the spanish word "libre", what means "freedom". And "gratis", what means "at no price". |
| irbis | It is more acurrate, and it will avoid misunderstanding. |
| irbis | It is suposed to beging introducing clustering technologies. |
| irbis | Anyway, all the years I begin introducing clustering. |
| irbis | That is why |
| irbis | I am not going to repeat again the talk of two years ago. It would consume my whole time, and It |
| irbis | can be read from umeet site. |
| irbis | If you do not know what is a cluster or what MOSIX means, I recommend reading the two years before talk. |
| irbis | My talk will be divided on three parts. |
| irbis | First one, is conceptual remark on what SSI really means. |
| irbis | Second one is an historical introduction to OpenMosix history, the causes of the fork and what OpenMosix has that MOSIX has not. |
| irbis | At last, I will divage on SSI, OpenMosix, and where the clustering world goes. |
| irbis | And I will finish with an insteresting section of questions and answers, and a little talk and roundtable. |
| irbis | I can respond questions in English, Spanish, and Portuguese. |
| irbis | Let´s go. |
| irbis | The SSI systems |
| irbis | Most of you have operated, or heared about clusters. |
| irbis | A cluster looks to the user and to the programmer like a cluster. |
| irbis | (A group of machines put together to do a determinate task) |
| irbis | This means that you have: |
| irbis | a) an special efford of the developers to develop parallel code |
| irbis | b) an special efford of the users -maybe de more paintfull of the problems- |
| irbis | c) an special efford to the sysadm |
| irbis | Then, you are changing user/developer/sysadm handwork by computational power |
| irbis | That did clusters grow on scientific area and at university |
| irbis | There, computational needs are really great, and labour is nearly "gratis" |
| irbis | (In fact, with undergraduate student, Ph. Dr. and other slave^M^M^M^Mstudent labour, you can |
| irbis | consider most of the labour completly free) |
| irbis | Anyway, the users are not happy. |
| irbis | Using a cluster is harder than a single machine, and needs skiller people than |
| irbis | a single machine. |
| irbis | What is the next step? |
| irbis | (I will slow down, the translator is being in trouble) |
| irbis | For most people, is buying a faster machine. |
| irbis | But there is time when you have not a faster monoprocesor machine. |
| irbis | That is when multiprocessor machines come at aid. |
| irbis | They are as simple to use and administrate as a monoprocessor machine. |
| irbis | But multiprocessor machines does not scale very well. |
| irbis | In fact, due to some memory and bus issues, upon a number of processors, it does not scale at all. |
| irbis | There it comes NUMA systems. |
| irbis | It is like multiprocessor architectures, but there is some differences. |
| irbis | Basically, each processor has its own local memory. |
| irbis | I am not going to lose too much time here. Our objetive is not a talk about NUMA. |
| irbis | NUMA has enormeous advantages versus normal machines: |
| irbis | Most of them are more powerful than monoprocessor and common multiprocessor machines. |
| irbis | It allows the play of the size: "mine is bigger than yours" |
| irbis | (This play is common between universities and departments) |
| irbis | It is a good choose if you need computational power, but it costs too much. |
| irbis | This is not problem if you has lots of bucks to throw away. |
| irbis | In fact, it has been choose by some throw away bucks with NUMA, and retain slave labour. |
| irbis | Then, we have: |
| irbis | monoprocessor has no enough power |
| irbis | multiprocessor has no enough power |
| irbis | NUMA is really expensive |
| irbis | clustering need skilled users, sysadms and developers |
| irbis | What would be the solution? |
| irbis | Something that can be used and programmed like a multiprocessor machine, |
| irbis | or like a NUMA machine, but really uses clustering hardware. |
| irbis | (here we have some easy jokes about Kinder (r) eggs) ;-) |
| irbis | It is the Saint Graal of supercomputing. |
| irbis | A cluster that behave like a single machine. |
| irbis | It is OpenMosix , a libre SSI clustering project |
| irbis | SSI remains for single system Image |
| irbis | The main idea is that the whole cluster looks like a single machine. |
| irbis | Imposible? |
| irbis | No. |
| irbis | There have been some succesful attemps on non-libre world. |
| irbis | And some attemps on Linux. |
| irbis | Now, the two only solutions that works are the "gratis" solution -MOSIX- and the "libre" solution -OpenMosix-. |
| irbis | First one, we had MOSIX. Two years ago I talked about MOSIX project. |
| irbis | We will back to it in few minutes. Because MOSIX works. |
| irbis | Second, we had SSI clustering atemp from Compaq. |
| irbis | Two years of vaporware, tons of people talking about it, nearly no code. |
| irbis | The few code of Compaq's SSI is "borrowed" from other libre software projects. |
| irbis | It is really incredible what money and PR departments can do. |
| irbis | Now, SSI does not do a "working platworm" |
| irbis | The third option is OpenMosix. |
| irbis | OpenMosix is a libre fork of MOSIX project. |
| irbis | OpenMosix is leaded by Dr. Moshe Bar. |
| irbis | And has some really well-skilled developers. |
| irbis | It has a bazaar-coding style. |
| irbis | To the user: |
| irbis | On OpenMosix, you need an administrator as skilled as needed to manage a normal cluster. |
| irbis | But the system looks to the programmer and to the user like a multiprocessor single machine. |
| irbis | That is SSI. |
| irbis | SSI exists. |
| irbis | It works. |
| irbis | And you can get it years ago. |
| irbis | {mode rant on} |
| irbis | Intentionaly, I have left industry out of all of this. |
| irbis | That is why, no matter I find really insteresting for the industry all this cluster |
| irbis | and SSI stuff, I find no interest at all in industry on some countries. |
| irbis | I have built some MOSIX and OpenMosix clusters at academia, bur I have find no interest at all |
| irbis | at any industry, neither Spain nor Brasil, to mount any OpenMoisix cluster. |
| irbis | Here, In Spain, I have OpenMosix like a hobby while searching for a new job. But |
| irbis | there is no commertial interest on it. Something really stupid, so far it is interesting |
| irbis | for the industry because it saves money. But it looks that Spain is too rich, and saving |
| irbis | money at IT on hardware or software is not an issue. :-( |
| irbis | {mode rant off} |
| irbis | Part II. An historical introduction to OpenMosix |
| irbis | In the beggining, it was MOSIX. |
| irbis | Since 70's MOSIX allows SSI pure clustering on POSIX environments, with transparent |
| irbis | task migration and load-balancing. |
| irbis | Anyway, the horrible PR policies of prof. Barak, the master genius behind MOSIX, |
| irbis | and the incredulity of the other researchers, did to MOSIX stay on the shadows during |
| irbis | 25 years. |
| irbis | When MOSIX was ported to Linux, due to the virical nature of the GPL, the kernel part |
| irbis | of MOSIX project had to be licenced under GPL. |
| irbis | A time after, due to some pressures, prof. Barak licence also the user area part of MOSIX |
| irbis | project -the so-called userland tools- to GPL. I know that prof. Barak personally thinks |
| irbis | that this was a great mistake. |
| irbis | During next years till 2001, MOSIX was a cathedral-style proyect, but its licence was libre and gratis. |
| irbis | Prof. Barak never acept any external help or patch. Opening the project gave him |
| irbis | lots of fame and the posibility of lots of bucks for development from IT industry. |
| irbis | But he is not a person fame-driven or money-driven. And he has Israel and US goverment |
| irbis | money to keep the project rolling. |
| irbis | This was the project of his life, and making it free, after two years, keeping the |
| irbis | control over MOSIX was becaming harder. |
| irbis | He rejected more industry aids and more people willing to help. |
| irbis | Meanwhile, the best skilled pupil of prof. Barak, Dr. Moshe Bar, became more and more |
| irbis | unhappier. |
| irbis | As we say in Spain "no puede haber dos gallos en un gallinero". |
| irbis | (Any help on the translation?) |
| irbis | Relations between prof. Barak and Moshe Bar became difficult. |
| irbis | One year ago, MOSIX project left the "libre" world. MOSIX now is gratis, but it is not libre. |
| irbis | In particular, they changed the licence of whatever they can to the "MOSIX free licence". |
| irbis | A "gratis" licence, but not a "free" licence. |
| irbis | This was the end of a story of misunderstanding and fights between prof. Barak, and Moshe Bar. |
| irbis | Dr. Moshe Bar now is the leader of OpenMosix project. And he has very good programers working on it. |
| irbis | {a short lines about prof. Barak} |
| irbis | I really admirate prof. Barak. |
| irbis | Prof. Barak is a genius. He is the original developer of load-balancing algorithms of MOSIX. |
| irbis | I know him in Paris two years ago. He was a very kind person, |
| irbis | and I leart more about clustering on two days with him than in four years on heavy interest |
| irbis | on clustering, developing for clusters and administrating clusters, including MOSIX clusters. |
| irbis | But the most important of this meeting was finding my own leaks. I spent the next year on my |
| irbis | free time reading the MOSIX source patch. |
| irbis | I must say this as a way of recognicement to the master, prof. Barak. |
| irbis | All the algoritms and the most briliant ideas developed in MOSIX are product of his research. |
| irbis | He had a very good team, including people like Oren La'andan and others. |
| irbis | Moshe Bar worked at Linux porting, and did the MFS -Mosix File Sytem- stuff. |
| irbis | But we must recognice the great scientific work of prof. Barak. |
| irbis | Anyway, the child of prof. Barak, MOSIX, had grown. |
| irbis | It was really big, and it has a skilled user base willing to help. |
| irbis | {hold on a minute. Translators need a little lag} |
| irbis | The fork between MOSIX and OpenMosix was too paintful. |
| irbis | There would be other ways to do it. |
| irbis | But it happend for the most paintfull way. |
| irbis | Kernel code was not specially hard to fork. |
| irbis | But in userland tools, things were harder. |
| irbis | I began with old code with lots of inocency |
| irbis | But Mulix, a very skilled programmer that deeply knows OpenMosix code said to me to keep on the eye the licence stuff. |
| irbis | Here I find that things were not so easy. |
| irbis | I had to return to a really old version to userland tools that I could prove in a tribunal that it were GPLed |
| irbis | and i had to throw away large junks of code, and some tools. |
| irbis | Anything that had not a "GPL" clause on the begining on the file were thrown away. |
| irbis | Personally, that was a disaster. |
| irbis | I was working on a new version of userland tools, completly rewritten and with aids to grid applications |
| irbis | and I had to stop the development and begin to patch a great package with lots of lines of code. |
| irbis | The first versions of the GPL-clean userland tools were for MOSIX and OpenMosix. |
| irbis | And I wanted the new userland tools that I was working were compatible with MOSIX and OpenMosix. |
| irbis | But after a few not friendly messages from prof. Barak, I decided to leave any compatibility on MOSIX. |
| irbis | I understand his position. MOSIX is the work of his life, and he thought that it went out of his direct control. |
| irbis | Anyway, science is built on the shoulders of gigants. But this gigant did not allowed anybody over his shoulders. |
| irbis | That means that any new code that I work for does not work under MOSIX. MOSIX is no more libre software, |
| irbis | and I am not more interesting on it. |
| irbis | From there, we hace evolved a lot from our source, MOSIX. |
| irbis | We have not done anything extrematelly briliant. We have done lots of things that it should be done, like bugfixing and |
| irbis | little features that help to the user and the administrator. |
| irbis | Basically, putting OpenMosix ready for enterprise. |
| irbis | Basically, putting OpenMosix ready for enterprise. |
| irbis | We have followed the kernel version -this wastes resources of the team, but must be done-. |
| irbis | TONS of bugfixes. On userland and in kernel land -in fact, I was acepted on the begining as colaborator |
| irbis | of OpenMosix project after it were acepted a deadlock fix on OpenMosix kernel part -MFS-. |
| irbis | New tools. Now, we have autodiscovery between nodes on a OpenMosix cluster. |
| irbis | And ALL the things that MOSIX has (plz look at two years ago presentation, where you can find all |
| irbis | the propieties common of a MOSIX/OpenMosix cluster) |
| irbis | Do not think in OpenMosix as in a revolution over MOSIX, but in an evolution. |
| irbis | Some cool new features, easier to configurate, some bugfixes, and it is libre. |
| irbis | Part III future works |
| irbis | We have now two main fronts of work: |
| irbis | Moshe and others are working on the port of OpenMosix to IA64 and following the kernel. |
| irbis | This is using the most of our resources. |
| irbis | HP gave machines for this, as far as I know. |
| irbis | Moshe has a firm, and is using it to do this stuff, and other OpenMosix-related stuff that I do not know. |
| irbis | Personally, I go at my own. |
| irbis | I do this on my free time, bug hunting userland code. |
| irbis | I am also working on my new OpenMosix userland tools. It keeps backwars compatibility, but it includes |
| irbis | a new library to do grid stuff on OpenMosix. |
| irbis | SSI for common processors + Grid for special processors + transparent process migration + load balancing is a winner |
| irbis | combiantion. |
| irbis | Common process migrate transparently to users and developers. |
| irbis | Grid process must be compiled against a special library, that allows breakpoints, multicasting on the same |
| irbis | function and result validation. But remote node execution, function migration and load balancing are also |
| irbis | transparent to the user. |
| irbis | My principal idea is to free the code as soon as it is on alpha stage and the paper were accepted. |
| irbis | Anyway, the development has slowed down due to my 14-hours laboral day on the first half of the year |
| irbis | and my job searching on the last months. |
| irbis | Anyway, if it is anybody wiling to know my opinion, grid is the future, and I am working on my free time to assure |
| irbis | that OpenMosix goes by there. |
| irbis | Well, this is a short introduction, following from where we stay two years ago. |
| irbis | Now, I think that a Q/A section, and a little round table would be great. Here casually is Mulix, |
| irbis | one of the persons that knows OpenMosix. In fact, he knows better of the born of OpenMosix and the relationships |
| irbis | between the players than me. |
| irbis | Thanks for reading me so much time. |
| mulix | clap clap clap clap clap lcap clap clap clap clap clap clap |
| mulix | clap clap clap clap clap lcap clap clap clap clap clap clap |
| mulix | thanks for the nice presentations, irbis |
| mulix | and for your work on the project. |
| irbis | Thanks, Mulix. |
| mulix | and thanks for the +o, sarnold :-) |
| irbis | Pask asked on #qc about "ha" and OpenMosix. |
| fernand0 | plas plas plas plas plas plas plas plas plas plas plas |
| fernand0 | ops |
| ShawnX | clap clap clap clap clap clap clap clap clap clap clap clap |
| fernand0 | plas plas plas plas plas plas plas plas plas plas plas |
| jsz | irbis: that was a very interesting speach |
| fernand0 | plas plas plas plas plas plas plas plas plas plas plas |
| fernand0 | plas plas plas plas plas plas plas plas plas plas plas |
| ShawnX | clap clap clap clap clap clap clap clap clap clap clap clap |
| visik7 | clap clap clap clap clap clap clap clap clap clap clap clap |
| ShawnX | clap clap clap clap clap clap clap clap clap clap clap clap |
| sarnold | irbis: thanks :) clap clap clap clap |
| sarnold | irbis: thanks :) clap clap clap clap |
| sarnold | irbis: thanks :) clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| irbis | Thanks. |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| ShawnX | clap clap clap clap clap clap clap clap clap clap clap clap |
| pask | clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap clap |
| pask | ;-) |
| fernand0 | plas plas plas plas plas plas plas plas plas plas plas |
| fernand0 | plas plas plas plas plas plas plas plas plas plas plas |
| MJesus | hummmm casually ? |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| murof | w zasadzie da³o siê zrozumieĉ :) |
| irbis | Well, about pask's questions. |
| irbis | Sorry, no polish on the chanel. |
| sarnold | dani, aka_mc2, thanks for providing translation :) |
| irbis | Sorry, no polish _language_ on the chanel. ;-) |
| irbis | Polish people are wellcome. :-) |
| jsz | (: |
| mulix | irbis, you mentiond grid computings... what kind of support do you think openMOSIX needs for grid computing? |
| jsz | irbis: we understand u |
| irbis | Well, Pask. If you do HA or fault tolerance on OpenMosix you must know exactly what is happening. |
| pask | the main problem in a vrrp ha firewalls is how to mantain the stateful tables between the nodes ... openmosix could help me |
| irbis | If a process migrates, and fails the home node or the destination node, you will lost the process. |
| mulix | openmosix isn't suited for high avaialability type of clusters, IMHO |
| irbis | Then, you have there an "anti-faulttolerance" |
| irbis | cluster. |
| pask | ... |
| ladypine | irbis, regarding the politics part: As I was told by a MOSIX user, the only reason MOSIX userland tools are not GLPed now, is the forking that Moshe Bar did, aiming to make money out of the academic work of years of others. |
| mulix | it's suited for HPC, where you have little IO, and can afford occasionally to lose a run (becauses you can repeat it) |
| pask | ok |
| irbis | Hold on a minute, ladypine. I will answer to you in a minute. Only response to pask before. |
| casanegra | clap clap clap clap clap clap clap clap clap clap |
| casanegra | clap clap clap clap clap clap clap clap clap clap |
| casanegra | clap clap clap clap clap clap clap clap clap clap |
| irbis | Ther eis a trik. If you can save the state of the process, and you have not problem if the process dies, |
| irbis | you can use an OpenMosix cluster to give masively a service. |
| irbis | In that case, but only in that case, you can use OpenMosix on a fault-tolerant escenario. |
| ShawnX | irbis: will OpenMOSIX be added to the Linux kernel? |
| pask | the problem is how to share and replicate /proc/net/ip_conntrack |
| ShawnX | or is it all userland |
| irbis | It is like doing HA with LVS: no matter that fails a single machine, you lose that process that were |
| pask | irbis ok .. |
| mulix | shawnx, there's a big kernel patch (50k diff, IIRC), which is very intrusive |
| mulix | unless it's rewritten, no chance, IMO. |
| ShawnX | mulix: perhaps for 2.7 :-) |
| irbis | generated on that machines, or run in that machines, but you does not loose the whole cluster. |
| ShawnX | i will accept it into the 2.6-pre kernel tree mulix. |
| irbis | As far as the control algorithm is completly distributed, |
| mulix | shawnx, didn't know you are the 2.6 maintainer :-) |
| ShawnX | no no |
| mulix | do you maintain a kernel tree? |
| sarnold | pask: well... dan kaminsky has a NAT tool that can be restarted and correctly pick up the state of current connections |
| mulix | you can get the patch from openmosix.sf.net |
| ShawnX | hightower and I will be maintaining a new tree an official 2.6 post tree |
| irbis | -one of the bigest ideas of prof. Barak- |
| sarnold | pask: maybe that could be made 'distributed' and help with connection tracking? |
| ShawnX | mulix: eXtendable Linux Kernel (-xlk) |
| pask | seth im not doing nat .. just forwarding |
| irbis | the fail of a node does not cause the stop od the cluster. |
| ShawnX | and we'll be accepting all kinds of experimental patches, including OpenMOSIX ;) |
| ShawnX | since it does belong in the kernel |
| irbis | Maybe mixing in some way LVS+Heartbeat+OpenMosix |
| ShawnX | mulix: its to get ready for the 2.7 dev tree. |
| pask | openmosix+vrrpv2 |
| irbis | I have not done it. |
| irbis | I do not know if it works. |
| pask | ok irbis .. |
| irbis | OpenMosix has a distributed control algorithm |
| irbis | no matter a node fails, the cluster continues working. |
| pask | ok but is the same /proc tree in all nodes? |
| irbis | I will look at it. Send to me a mail (irbis@orcero.org) |
| ShawnX | mulix: more news on this kernelset in 5-6 months :-) |
| irbis | and we can look at it carefully. |
| pask | ok thanks at all |
| sarnold | irbis: how much actual coding does moshe bar do these days? I've read some of his articles and he scares me :) |
| irbis | (response to ladypine) |
| mulix | sarnold, as far as I can see from the maiilng list and CVS, very little to none. |
| irbis | Ladypine, it is not as easy as it looks. |
| jsz | irbis: u are working in kernel project ? |
| irbis | I have not earn a dolar with all this stuff. |
| irbis | And I maintain the userland tools. |
| irbis | What prof. Barak did is to lost the bosibility of use my future work and improvements. |
| uninet | how does mosix react to network partitioning? is there the concept of primary partition? |
| irbis | In fact, I have been wasting my time repeating things that have be done by MOSIX on a completly different way, |
| irbis | for not to break the law. |
| ladypine | irbis, I never said you make money of that. I said Bar intends to make money from his company, Qlusters, which deals with distributed operating systems. |
| garoeda | clap clap clap clap clap clap |
| garoeda | clap clap clap clap clap clap |
| garoeda | clap clap clap clap clap clap |
| garoeda | clap clap clap clap clap clap |
| irbis | If I had no lost that time repeating code by the stupid war, MOSIX and OpenMosix would have now |
| irbis | a library with grid semantics. |
| murof | what is clap clap?? |
| irbis | Now, none of them have it. |
| irbis | It is not briliant. |
| garoeda | mulix: applaus |
| mulix | murof, it means that people are clapping for the speaker. |
| sarnold | murof: applause |
| MJesus | aplause murof |
| irbis | I admirate deeply prof. Barak, but he was no clever on this desition. |
| ladypine | irbis, BTW, I was wondering for quite some time where is Bar's PhD from. I never managed to find any thesis by him. |
| murof | oh i understand: clap clap clap :) |
| murof | clap clap |
| irbis | ladypine, please repeat the coment before the last, too much conversation on the channel. |
| ladypine | irbis, I never said you make money of that. I said Bar intends to make money from his company, |
| ladypine | Qlusters, which deals with distributed operating systems. |
| irbis | ladypine, I have not readed it also. I does not ask for a copy of the thesis of each person that says that has a Dr. |
| irbis | As far as I can renember, I have seen this on some presentations. Maybe I am wrong, and I never had heared it. |
| irbis | Maybe you are wrong, and he has the degree. |
| irbis | About Qlusters. |
| mulix | irbis, don't you find it strange that someone calls himself a Dr. but his thesis cannot be found anywhere? (I looked for it as well). |
| irbis | There is lots of people that does not have his PhDr on Internet. |
| irbis | I does not need to suspect from anybody I find. I enyoy coding, and I try to do reseach on grid over the OpenMosix stuff. |
| irbis | It is all about this. |
| irbis | Returning to Qlusters. |
| mulix | irbis, anyway, forget about it, let's talk about the openmosix project |
| irbis | I do not find anything bad about Moshe having a firm. |
| Vegas | hi epic |
| irbis | IBM earns money in some countris with MOSIX, and I doubt that prof. Barak be agry with |
| irbis | them. |
| mulix | irbis, do you have any details on that? |
| irbis | On Qlusters, I do know nothing. It is a Moshe thing, and I do not know anything about it, and we have not talked about it. |
| irbis | About IBM, yes. He is mounting MOSIX clusters in France. |
| mulix | irbis, never mind Moshe or Qlusters, let's talk about the code :-) |
| sarnold | i wonder why discussions of mosix always tend to lead to stories about moshe... |
| irbis | There is also other firms that sells MOSIX clusters, |
| irbis | in fact I present prof. Barak and one of that firms in Paris, two years ago, and he did not looked disapointed. |
| mulix | sarnold, something is rotten in the kingdom of Denmark. |
| irbis | Well, Mulix, I would like about the code |
| irbis | so far there are interesting things to talk about. |
| sarnold | Arador: buenas noches |
| mulix | irbis, do you know if there are any plans for a rewrite of the code, to make integrate it into the Linux kernel? |
| mulix | (no way that it will get accepted in its current state) |
| Arador | hola sarnold ;) |
| irbis | Well, this is very complex. And I should before to address something that you yalked with Moshe on the mailing list. |
| mulix | the issue of stability? |
| irbis | The issue of the patch. |
| mulix | do you agree with me or Moshe? |
| mulix | two people, four opinions, as someone said :-) |
| irbis | Part of each one. |
| irbis | Three people, three opinions. |
| mulix | which part do you disagree with? |
| irbis | By a side, OpenMosix touches some things that are not a module on Linux. |
| irbis | Some parts, as MFS, are at a alpha stage, and I am not sure that it would be ready for enterprise. |
| irbis | University is one thing, enterprise is another. |
| irbis | If you would put the database of your firm where it is the core of your business, the filesystem is ready |
| irbis | but i would not do this, after reading parts of the code. |
| mulix | judging by the recent bugs in MFS and openmosix in general, I don't think it's ready. |
| irbis | Then, the parte that is more capable of runing as module, is in alpha stage. |
| garoeda | MJesus: and another fine translation done |
| irbis | By the other side, the part that is hard-coded on critical parts is stable. |
| irbis | (or, as stable as Linux is. With some deadlocks, but better than Windows 98) |
| irbis | The most of the resources are wasted folowing the kernel. |
| irbis | It is a running target. |
| MJesus | garoeda thanks!! |
| irbis | And we can not test it throughtly -you are righ here-. |
| mulix | irbis, in that case, why not stop following the kernel and concentrate on features and stability? |
| irbis | Good question. |
| sarnold | mulix: because resyncing with the kernel many revisions later is _very_ difficult, and probably more expensive than tracking many smaller changers |
| irbis | IMHO, we should break down OpenMosix on some different patches. |
| irbis | If we want to be in the kernel some day, we should do this. |
| hightower | irbis: that would be perfect! |
| mulix | sarnold, then again, if a project spends all of its man power tracking kernel revisions, it is either doing something wrong or has the wrong people. |
| mulix | irbis, the code is way too un-modular to be able to do this right now... |
| mulix | IMO, opinion. |
| sarnold | mulix: I understand the choice of tools makes a huge difference.. doing it by hand is evidently many times slower than using bitkeeper's three-way merge tool |
| taigeru | or has less manpower than it should... |
| irbis | But the core OpenMosix part is unmodular, you are right, and it is not clear that some part could be modular, there Moshe is right. |
| mulix | taigeru, or that, too |
| mulix | irbis, I never said that some parts could be modular in the current design - I said that some parts need to be *rewritten* to be modular. |
| irbis | Nonono... it is not a OpenMosix problem, is a Linux kernel problem. |
| sarnold | irbis: note that people who think that way tend to never get their patches accepted :) |
| irbis | You can not load a new scheduler as a module. |
| mulix | irbis, you're thinking about it the wrong way - opebnmosix should *not* just insert its code into the scheduler |
| mulix | it should get suitable hooks or patches in the main kernel, and use those to make its changes non intrusive |
| irbis | Sarnold, you are parcially right, but the nature of OpenMosix is special because it touches nearly everithing on the kernel process subsystem. |
| irbis | We need do some things: |
| irbis | first, taking away al MFS stuff as a separate patch. |
| mulix | irbis, when I offered to do things in the Linux kernel way, Moshe objected to the patch because "he doesn't like it", with no other expalantion |
| irbis | second, using hooks to throw to userland so may things that we could. |
| mulix | even though what he suggested has no change in hell of ever being accepted (ifdefs for logging, for example) |
| mulix | irbis, we agree on "do everything you can in userspace" part |
| irbis | third, using hooks for that things that were needed. The problem is that I see too many hooks, and the solution would be a process subsystem more modular. |
| irbis | But this is not the worst. |
| irbis | If we have had ony one SSI clustering project, we would finish in the kernel sooner or later. |
| irbis | But we have THREE SSI clustering projects. OK, Compaq SSI does not work, but it does not matter. |
| irbis | I have read some important kernel hackers talk well about this project, no matter they have not tested it. |
| irbis | And we have TWO SSI projects that works. |
| irbis | IMHO, the fork made nearly impossible in the near future to include ourselves on the mainline. It give to us a lesser userbase, |
| irbis | and a weaker position on any talk with kernel people. |
| sarnold | imho, no matter forks etc, if one codebase is in mergable condition, and the other not, the fork doesn't much matter |
| MJesus | the last day, deember 20, there are a round table with 5 kernel hackers |
| mulix | sarnold, teh MOSIX and openmosix kernel code are 99% identical |
| mulix | and neither is (IMO) mergable. |
| irbis | But now we are not remotely mergeable. |
| mulix | mjesus, who are the kernel hackers? |
| irbis | MOSIX and OpenMosix will not merge due to personal issues |
| irbis | and OpenMosix and Linux can not merge due to technical issues |
| mulix | irbis, MOSIX cannot be merged due to teh same technical issues |
| mulix | I remember Al Viro looking at the openmosix code for less thana minute and finding a huge race ;-) |
| irbis | no, I am saying: MOSIX and OpenMosix will not merge again between them due to personal issues. |
| sarnold | mulix: riel (if he can make it) dave jones, robert love, akpm, bacchus |
| irbis | You know beter than me what happened. |
| irbis | And we must recode some parts and redesign others to make OpenMosix mergeabe. |
| MJesus | and sarnold |
| mulix | irbis, I know, but some of that knowledge is confidential and I cann;t talk about it. |
| jsz | KVIrc 2.1.2 'Monolith' |
| irbis | It is a slow process, and OpenMosix should have to merge step by step, not ALL openmosix in a time. |
| mulix | sarnold, coolness... I'll be sure to be here. |
| mulix | sarnold, who is bacchus IRL? |
| sarnold | mulix: ralf bachle (with umlaut over the a in bachle :) |
| irbis | OK, Mulix. I know that I do not know what _really_ happened. But I know what happened to me. That was I talked, what I lived. |
| mulix | sarnold, oh, 'k. |
| hightower | sarnold: just type an ä |
| hightower | sarnold: ;) |
| sarnold | hightower: i'd love to :) but I didn't want to take the time to find it :) |
| mulix | irbis, let's just agree that it's a complex story and no one is telling the whole truth, as far as I know. |
| hightower | hehehe |
| mulix | irbis, how is your work search going? |
| sarnold | Tue Dec 17 20:07:48 UTC 2002 |
| irbis | Anyway, I thing that the part of merging Linux and OpenMosix is really interesting, |
| mulix | irbis, I'm sorry, but I'll have to leave the office and finally go home now, it's after 10 PM here |
| irbis | I feel that all that discussion about modularizin and how to modularize is a discussion that all we have to had. |
| sarnold | mulix: good night :) |
| irbis | OK, I undersand. We soulkd keep in the future to talk. |
| mulix | thank you for the talk, and I'll see you on the openmosix mailing lists :-) |
| irbis | Good night. |
| MJesus | and thanks you all ! |
| mulix | sarnold, everyone, good night. |
| jsz | bye irbis |
| irbis | bye. |
| aka_mc2 | adios |
| aka_mc2 | david |
| aka_mc2 | seguimos en contacto por emial, ultimemente no te llegan... |
| aka_mc2 | seguimos en contacto por email, ultimamente no te llegan... |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| MJesus | clap clap clap clap clap clap clap clap clap clap |
| jsz | next speach at 22.00 |
| hightower | thank you irbis! |
| sarnold | irbis: thanks :) |
| irbis | Thanks to all. |
| aka_mc2 | a ti |
| aka_mc2 | ! |
| aka_mc2 | :D |
| irbis | Thanks to you. |