Menu
▾
▴
ci-linux-devel — Discussion about Cluster Infrastructure for Linux
You can subscribe to this list here.
| 2001 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
(25) |
Jul
(13) |
Aug
(11) |
Sep
(14) |
Oct
(5) |
Nov
(7) |
Dec
(4) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2002 |
Jan
(14) |
Feb
(10) |
Mar
(30) |
Apr
(9) |
May
(20) |
Jun
(12) |
Jul
(7) |
Aug
(6) |
Sep
(6) |
Oct
(34) |
Nov
(14) |
Dec
(9) |
| 2003 |
Jan
|
Feb
(9) |
Mar
(2) |
Apr
(2) |
May
(5) |
Jun
(14) |
Jul
(1) |
Aug
(7) |
Sep
(6) |
Oct
(5) |
Nov
|
Dec
|
| 2004 |
Jan
|
Feb
(1) |
Mar
(1) |
Apr
|
May
|
Jun
(1) |
Jul
(2) |
Aug
(2) |
Sep
(4) |
Oct
|
Nov
(7) |
Dec
(1) |
| 2005 |
Jan
(1) |
Feb
(2) |
Mar
|
Apr
(4) |
May
(5) |
Jun
|
Jul
|
Aug
|
Sep
(1) |
Oct
|
Nov
|
Dec
|
| 2006 |
Jan
|
Feb
|
Mar
|
Apr
|
May
(11) |
Jun
(2) |
Jul
|
Aug
(5) |
Sep
(5) |
Oct
(1) |
Nov
(1) |
Dec
|
| 2007 |
Jan
|
Feb
(1) |
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2008 |
Jan
(2) |
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2009 |
Jan
|
Feb
|
Mar
|
Apr
|
May
(1) |
Jun
|
Jul
|
Aug
|
Sep
|
Oct
(2) |
Nov
|
Dec
|
| 2011 |
Jan
(1) |
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
|
From: Brian J. W. <Bri...@Co...> - 2001-08-03 20:24:27
|
--
Brian Watson | "The common people of England... so
Linux Kernel Developer | jealous of their liberty, but like the
Open SSI Clustering Project | common people of most other countries
Compaq Computer Corp | never rightly considering wherein it
Los Angeles, CA | consists..."
| -Adam Smith, Wealth of Nations, 1776
mailto:Bri...@co...
http://opensource.compaq.com/
|
|
From: <kva...@ya...> - 2001-07-30 18:53:57
|
Hi , Points 1,2 remains same 3) edit your aboot.conf file eg: sample entry in aboot.conf 3:5/vmlinux-cluster root=/dev/sda2 CLUSTER_NODENUM=2 NOTE There is no "append" here 4) make dep vmlinux modules 5) gzip vmlinux it will create vmlinux.gz . 5) cp vmlinux.gz /boot/vmlinux-cluster make modules_install 6) create /etc/lilo.conf and necessary entries as specified in the other documentation 7) reboot At the SRM prompt give boot -fl 3 ( note the I used the number '3' because the sample entry of aboot.conf I quoted above had have '3' as the number . -aneesh ____________________________________________________________ Do You Yahoo!? Get your free @yahoo.co.uk address at http://mail.yahoo.co.uk or your free @yahoo.ie address at http://mail.yahoo.ie |
|
From: Brian J. W. <Bri...@co...> - 2001-07-25 05:43:40
|
The slides that were previously on the SSI website were given to me by Bruce before he left for the Montreal Linux Expo. The slides that are now up are what he actually presented. They're changed somewhat (there's five less slides), and the template complies with the Compaq corporate standard (oh boy). More importantly, Bruce recently wrote annotation for almost every slide. To view the slides, point your browser at: http://bjbrew.org/cpq/ssic_linux/montreal/index.htm -- Brian Watson | "The common people of England... so Linux Kernel Developer | jealous of their liberty, but like the Open SSI Clustering Project | common people of most other countries Compaq Computer Corp | never rightly considering wherein it Los Angeles, CA | consists..." | -Adam Smith, Wealth of Nations, 1776 mailto:Bri...@co... http://opensource.compaq.com/ |
|
From: Bruce W. <br...@ka...> - 2001-07-20 03:03:45
|
> > What do they look like? Are they a wide integer, or a structure, > or what? The initial cut is simple: a. you can optionally configure them on b. you set your node number in lilo c. MAXPID is changed to be the full 31 bits d. low order 16 bits are uniquifier e. high order 15 bits are node number f. no new apis (pids really are 32 bits in Linux) g. no known problems with utilities or apps or any such Where could we go from here: a. make the config stuff more flexible and let you specify how many bits for node number and how many for uniquifier (then you could fit in 16 bits if you really wanted to) b. make the whole cluster configuration more filesystem or GUI or whatever. I have no plans to do either of these but would welcome others who want to. bruce > > > > > br...@ka... wrote: > > > > There is a new release of Cluster Infrastructure. Besides moving > > up to 2.4.6 kernel and including > > the patches to allow it run on Alpha, it has support for > > clusterwide pids (we brought it over from the SSI project > > since there seemed to be interest). In doing the configuration > > of the kernel, the CPID capability is optional. > > > > So far the configuration capability is pretty simple. If you have > > a hankering to tinker, feel free. > > > > bruce walker > > > > _______________________________________________ > > ci-linux-devel mailing list > > ci-...@op... > > http://www.opensource.compaq.com/mailman/listinfo/ci-linux-devel > > -- > David Nicol 816.235.1187 > "Your plastic pal who's fun to be with" > |
|
From: Peter B. <tab...@ya...> - 2001-07-19 23:22:04
|
--- br...@ka... wrote: > > There is a new release of Cluster Infrastructure. Besides moving > up to 2.4.6 kernel and including > the patches to allow it run on Alpha, it has support for > clusterwide pids (we brought it over from the SSI project > since there seemed to be interest). In doing the configuration > of the kernel, the CPID capability is optional. Thanks, I'm trying to get around to giving this a shot on my machines. I would like to make sure I explicitly ask you keep the CPID capability optional in the CI code. I've no objection to it on the SSI side, but I don't want it in my CI clusters (maybe in the future, but, don't know that far ahead yet.) > > So far the configuration capability is pretty simple. If you have > a hankering to tinker, feel free. > > > bruce walker Peter ===== These have been the opinions of: Peter R. Badovinatz -- (503)578-5530 (TL 775) wo...@us.../tab...@ya... and in no way should be construed as official opinion of IBM, Corp. __________________________________________________ Do You Yahoo!? Get personalized email addresses from Yahoo! Mail http://personal.mail.yahoo.com/ |
|
From: <kva...@ya...> - 2001-07-19 15:15:40
|
Hi ,
Today I build the kernel with CPID enabled on
alpha. It seems it is working . I haven't faced any
issue till now . I did the following modification
to the files
arch/alpha/config.in
376 comment 'Internode Communication Subsystem
(ICS) is required for CLMS'
377 define_bool CONFIG_ICS y
378 fi
379 bool 'Clusterwide Pids' CONFIG_CPID
380 fi
The number indicate the line number 642
CONFIG_CLUSTER=y
arch/alpha/defconfig
643 CONFIG_CLMS=y
644 CONFIG_ICS=y
645 # CONFIG_CPID is not set
Since it is a one line change I am not sending the
patch. If anyone is interested I can generate the
patch and send it across
-aneesh
____________________________________________________________
Do You Yahoo!?
Get your free @yahoo.co.uk address at http://mail.yahoo.co.uk
or your free @yahoo.ie address at http://mail.yahoo.ie
|
|
From: <kva...@ya...> - 2001-07-19 11:01:19
|
Hi , include/asm directory in cluster tools contain unistd.h file specific to i386. So build of clustertools on alpha will fail. To correct this either copy unistd.h coresponding to alpha from cluster infrasture or have a directory structure simillar to linux kernel where asm is a link to platform specific directories. -aneesh ____________________________________________________________ Do You Yahoo!? Get your free @yahoo.co.uk address at http://mail.yahoo.co.uk or your free @yahoo.ie address at http://mail.yahoo.ie |
|
From: <br...@ka...> - 2001-07-19 07:04:50
|
There is a new release of Cluster Infrastructure. Besides moving up to 2.4.6 kernel and including the patches to allow it run on Alpha, it has support for clusterwide pids (we brought it over from the SSI project since there seemed to be interest). In doing the configuration of the kernel, the CPID capability is optional. So far the configuration capability is pretty simple. If you have a hankering to tinker, feel free. bruce walker |
|
From: Brian J. W. <Bri...@co...> - 2001-07-10 10:31:53
|
We would like to get some feedback on which kernel versions we should
release CI patches against. Right now we have patches for vanilla
2.4.2 and 2.4.4. Very soon, we're planning to release CI 0.5.6, which
will support Alpha (in addition to x86) and include some bug fixes.
For that release, we would like to bring the patch up to 2.4.6 and
just release for that kernel version. Are there any objections?
After that we're thinking of doing each CI release on a single recent
version of the kernel. The version of the kernel we do it on will not
always be the latest, depending on how many cycles we can spare to do
the merge. Of course, we wouldn't turn down a volunteer who wanted to
do the merge instead. ;)
Once 2.4 has stabilized and been handed off to Alan Cox, it may make
sense to support more than one kernel version. At that point, people
might be using old kernels because other patches they need haven't
been updated in awhile. Until then, however, I'm guessing that anybody
who knows how to build a kernel wouldn't mind being on one of the
latest and (hopefully) least buggy bases.
Please let us know what you think.
--
Brian Watson | "The common people of England... so
Linux Kernel Developer | jealous of their liberty, but like the
SSI Clustering Laboratory | common people of most other countries
Compaq Computer Corp | never rightly considering wherein it
Los Angeles, CA | consists..."
| -Adam Smith, Wealth of Nations, 1776
mailto:Bri...@co...
http://opensource.compaq.com/
|
|
From: Brian J. W. <Bri...@co...> - 2001-07-10 09:27:52
|
> We are doing a study research on the cluster > Infrastructure and SSI. Since we have gone thrrough > the kernel patch given in official compaq site, we > have problems compiling it for Redhat Linix Release > 7.1, Kernel 2.4.2-2smp on a 2-processor i686. > > The problem is, the kernel include files does not > contain "rwsem.h". Our trys to locate it in the net > are in vain. Official Redhat site is not providing the > patch for the same. You're applying a patch meant for 2.4.4 against a 2.4.2 kernel. Either get a 2.4.4 kernel from kernel.org, or download our 2.4.2 patch: http://www.linuxalpha.compaq.com/ci-linux/files/ci-linux-2.4.2-v0.5.5.tar.bz2 Please note that we've only tested this patch against vanilla 2.4.2, and not Red Hat's modified 2.4.2-2 kernel. Nevertheless, it'll probably work on Red Hat's kernel. > > Is there any special hardware required to make the > cluster? (As ServeNet PCI Adapter required for a 2 to > 6 node Proliant Clusters on UW7). > Nah. These days we're using run-of-the-mill ethernet as our interconnect. -- Brian Watson | "The common people of England... so Linux Kernel Developer | jealous of their liberty, but like the SSI Clustering Laboratory | common people of most other countries Compaq Computer Corp | never rightly considering wherein it Los Angeles, CA | consists..." | -Adam Smith, Wealth of Nations, 1776 |
|
From: Brian J. W. <Bri...@co...> - 2001-07-06 04:07:37
|
akash ghate wrote: > > Hi Folks, > I tried to compile the linux kernel with the > Clister Insfrastructure patch. > I got these errors. > Please help in compiling the code. > Thanks in advance. > Amol Ghate. > > In file included from > /usr/src/linux/include/cluster/ics.h:24, > from > /usr/src/linux/include/cluster/clms.h:28, > from clms_api.c:27: > /usr/src/linux/include/cluster/synch.h:28:25: > linux/rwsem.h: No such file or directory Parsing your errors, it looks like you applied the 2.4.4 patch against a 2.4.2 kernel. The big difference for us between the two kernel versions is how read-write semaphores are implemented. -- Brian Watson | "The common people of England... so jealous Linux Kernel Developer | of their liberty, but like the common Compaq Computer Corp | people of most other countries never Los Angeles, CA | rightly considering wherein it consists..." 310-414-3119 | -Adam Smith, Wealth of Nations, 1776 |
|
From: Brian J. W. <Bri...@co...> - 2001-07-06 03:49:23
|
> Hi , For building cluster tools for alpha please remove unistd.h file from cluster-tools-0.5.5/include/asm Procedure for applying the patch 1) Download kernel version 2.4.2 from alphalinux.org 2) apply the Cluster infrastructure patch from http://opensource.compaq.com/sourceforge/project/?group_id=22 3) Apply the following patch If there is any issue please let me know. -aneesh |
|
From: Brian J. W. <Bri...@co...> - 2001-07-06 03:48:30
|
Hi Brian , I got both the alpha - nodes up with the latest patch. There are something i noticed in the latest codebase 1) missing /etc/rc.nodeup ( The cluster_start is trying to execute it but the file is missing. even rc.nodedown is copied /etc/rc.d/ ) 2 ) the unistd.h in cluster-tools/include/asm directoy . Since this is a platform dependent file and linux have different file for different hardware ( like one for i386 and other for alpha ) it is good to make the compiler pick the file from linux kernel tree rather than providing it along with cluster tools . 3) redifing __NR_ssisys in ssisys.c file I have a different value for alpha and I have it defined it in asm/unistd.h. So it will be good to pick it from there . I will send you the patch one I clean the code base -aneesh "Brian J. Watson" wrote: > CI 0.5.5 and Cluster Tools 0.5.5 are now available. As I said a couple > of days ago, the primary change is in how a node joins a cluster. > Previously, it joined before mounting the root and got its information > through LILO. Now it joins when a command is run (sometime after > mounting the root), and it gets its information from a configuration > file. > > The source is available on our project summary page: > > http://opensource.compaq.com/sourceforge/project/?group_id=22 > > Hopefully we'll get CVS up soon, so that those interested in the > bleeding edge won't have to wait for a release. We have a few issues > to work through first. > > -Brian |
|
From: Brian J. W. <Bri...@co...> - 2001-07-06 03:31:59
|
"N, Ramesh" wrote: > > we are not able to send mails to your domain from xko.dec.com > > any reasons? > Aneesh Kumar's been having trouble posting from a digital.com email address. It looked like the problem in his case was in his M$ Exchange server. He said his MIS people are looking at it. I don't know if the problem's been resolved. opensource.compaq.com servers are not under our control. I don't know if there's something funky with their setup that's causing problems. -- Brian Watson | "The common people of England... so jealous Linux Kernel Developer | of their liberty, but like the common Compaq Computer Corp | people of most other countries never Los Angeles, CA | rightly considering wherein it consists..." 310-414-3119 | -Adam Smith, Wealth of Nations, 1776 |
|
From: David B. Z. <dav...@co...> - 2001-06-30 00:51:14
|
Are you using 0.5.0 or 0.5.5? And which linux kernel 2.4.2 or 2.4.4? akash ghate wrote: > Hi Folks, > I tried to compile the linux kernel with the > Clister Insfrastructure patch. > I got these errors. > Please help in compiling the code. > Thanks in advance. > Amol Ghate. > > In file included from > /usr/src/linux/include/linux/raid/md.h:51, > from init/main.c:25: > /usr/src/linux/include/linux/raid/md_k.h: In function > `pers_to_level': > /usr/src/linux/include/linux/raid/md_k.h:39: warning: > control reaches end of non-void function > In file included from > /usr/src/linux/include/cluster/ics.h:24, > from > /usr/src/linux/include/cluster/clms.h:28, > from clms_api.c:27: > /usr/src/linux/include/cluster/synch.h:28:25: > linux/rwsem.h: No such file or directory > In file included from > /usr/src/linux/include/cluster/ics.h:25, > from > /usr/src/linux/include/cluster/clms.h:29, > from clms_api.c:28: > /usr/src/linux/include/cluster/synch.h: In function > `down_write_trylock': > /usr/src/linux/include/cluster/synch.h:110: > `RWSEM_ACTIVE_WRITE_BIAS' undeclared (first use in > this function) > /usr/src/linux/include/cluster/synch.h:110: (Each > undeclared identifier is reported only once > /usr/src/linux/include/cluster/synch.h:110: for each > function it appears in.) > /usr/src/linux/include/cluster/synch.h:110: conversion > to non-scalar type requested > /usr/src/linux/include/cluster/synch.h: In function > `down_read_trylock': > /usr/src/linux/include/cluster/synch.h:121: aggregate > value used where an integer was expected > /usr/src/linux/include/cluster/synch.h:125: conversion > to non-scalar type requested > make[3]: *** [clms_api.o] Error 1 > make[2]: *** [first_rule] Error 2 > make[1]: *** [_subdir_clms] Error 2 > make: *** [_dir_cluster] Error 2 > > __________________________________________________ > Do You Yahoo!? > Get personalized email addresses from Yahoo! Mail > http://personal.mail.yahoo.com/ > _______________________________________________ > ci-linux-devel mailing list > ci-...@op... > http://www.opensource.compaq.com/mailman/listinfo/ci-linux-devel -- David B. Zafman | Compaq Computer Corporation mailto:dav...@co... | http://www.compaq.com "Thus spake the master programmer: When you have learned to snatch the error code from the trap frame, it will be time for you to leave." |
|
From: David B. Z. <dav...@co...> - 2001-06-30 00:48:06
|
This is NOT Alpha specific. We disable all non-init SSI system calls until the initialization of clustering is complete. In this case we could have allow clusternode_num() system call to function once the cluster_config --prep has completed successfully. Otherwise, we need to fix cluster_start. Aneesh Kumar wrote: > Hi , > > While working with the latest changes with > CI-Linux I noticied that clusternode_num return error > saying " This command require a Nsc kernel" if you run > it before cluster_config --postroot . This happens on > Alpha Linux I haven't checked it for Intel boxes. If > this is specific to Alpha please let me know :) . If > not I think cluster_start need some modifciation. You > will have to move the line that figure out the node > number to somewhere after the cluster_config > --postroot is done. .. > > I was not able to send the CI-Linux patch (wrt Alpha) > to the mailing list since this list is not accesible > from my company account . I have already send the > patch to Brian and Kai. Sorry :) > > -aneesh > > ____________________________________________________________ > Do You Yahoo!? > Get your free @yahoo.co.uk address at http://mail.yahoo.co.uk > or your free @yahoo.ie address at http://mail.yahoo.ie > _______________________________________________ > ci-linux-devel mailing list > ci-...@op... > http://www.opensource.compaq.com/mailman/listinfo/ci-linux-devel -- David B. Zafman | Compaq Computer Corporation mailto:dav...@co... | http://www.compaq.com "Thus spake the master programmer: When you have learned to snatch the error code from the trap frame, it will be time for you to leave." |
|
From: N, R. <Ram...@co...> - 2001-06-29 13:35:13
|
we are not able to send mails to your domain from xko.dec.com any reasons? Warm regards Ramesh N |
|
From: <kva...@ya...> - 2001-06-29 12:43:21
|
Hi , While working with the latest changes with CI-Linux I noticied that clusternode_num return error saying " This command require a Nsc kernel" if you run it before cluster_config --postroot . This happens on Alpha Linux I haven't checked it for Intel boxes. If this is specific to Alpha please let me know :) . If not I think cluster_start need some modifciation. You will have to move the line that figure out the node number to somewhere after the cluster_config --postroot is done. .. I was not able to send the CI-Linux patch (wrt Alpha) to the mailing list since this list is not accesible from my company account . I have already send the patch to Brian and Kai. Sorry :) -aneesh ____________________________________________________________ Do You Yahoo!? Get your free @yahoo.co.uk address at http://mail.yahoo.co.uk or your free @yahoo.ie address at http://mail.yahoo.ie |
|
From: akash g. <agh...@ya...> - 2001-06-29 04:52:47
|
Hi Folks,
I tried to compile the linux kernel with the
Clister Insfrastructure patch.
I got these errors.
Please help in compiling the code.
Thanks in advance.
Amol Ghate.
In file included from
/usr/src/linux/include/linux/raid/md.h:51,
from init/main.c:25:
/usr/src/linux/include/linux/raid/md_k.h: In function
`pers_to_level':
/usr/src/linux/include/linux/raid/md_k.h:39: warning:
control reaches end of non-void function
In file included from
/usr/src/linux/include/cluster/ics.h:24,
from
/usr/src/linux/include/cluster/clms.h:28,
from clms_api.c:27:
/usr/src/linux/include/cluster/synch.h:28:25:
linux/rwsem.h: No such file or directory
In file included from
/usr/src/linux/include/cluster/ics.h:25,
from
/usr/src/linux/include/cluster/clms.h:29,
from clms_api.c:28:
/usr/src/linux/include/cluster/synch.h: In function
`down_write_trylock':
/usr/src/linux/include/cluster/synch.h:110:
`RWSEM_ACTIVE_WRITE_BIAS' undeclared (first use in
this function)
/usr/src/linux/include/cluster/synch.h:110: (Each
undeclared identifier is reported only once
/usr/src/linux/include/cluster/synch.h:110: for each
function it appears in.)
/usr/src/linux/include/cluster/synch.h:110: conversion
to non-scalar type requested
/usr/src/linux/include/cluster/synch.h: In function
`down_read_trylock':
/usr/src/linux/include/cluster/synch.h:121: aggregate
value used where an integer was expected
/usr/src/linux/include/cluster/synch.h:125: conversion
to non-scalar type requested
make[3]: *** [clms_api.o] Error 1
make[2]: *** [first_rule] Error 2
make[1]: *** [_subdir_clms] Error 2
make: *** [_dir_cluster] Error 2
__________________________________________________
Do You Yahoo!?
Get personalized email addresses from Yahoo! Mail
http://personal.mail.yahoo.com/
|
|
From: Ramachandra P. <lin...@ya...> - 2001-06-28 04:54:53
|
Bruce, Well I am working on clustering aspects of EVMS(Enterprise Volume management system). As part of volume management startegy we have to make sure that every node in the cluster having concurrent access to volumes and volume groups sees the same state. This essentially leads to a requirement that there is many to many communication between nodes. (unless we have a master slave model; which I am exploring). The point here is everybody should see the same sequence of messages, sent by anybody; for everybody to build a consistent view of the system on all the nodes. For such a requirement, we could rely on techniques like globally ordered reliable messaging. Send-response trasactional model is also one of them, but isn't it applicable only in master-slave kind of scenarios? I believe there are other techniques to accomplish the same thing. I wonder, if DLM also requires this feature; though I am not positive. Thanks, Ram Pai -- Bruce Walker <br...@ka...> wrote: > Ram, > Thanks for the interest. > We do not have specific support for global > reliable ordered messaging. The reason > is that in all our years of SSI kernel > clustering, we have rarely needed it. > Algorithms that require talking to all nodes > in the cluster do not scale well and we > architectually avoid them whenever possible. > > One exception to that is the Cluster Membership > Service. The current master node must step > all the nodes through a couple of steps when > nodes come up or down. To do that, we tend to > use the request/response (also knows as async > RPC) feature of ICS. Using the interface to the > CLMS, a subsystem can determine the nodes to send > a given message to. He then can send them all > (no interface in ICS to sendall so this is a call > per node), and then wait for all the responses, > after which he can go on to the the next phase. > > Are there specific algorthims that you feel need > this > kind of capability. If so, maybe we can discuss the > need and maybe add this kind of capability. > > bruce > > > Hi, > > Trying to educate myself on CI. > > > > Looking at the features published on the web > page, > > I did not find any support for global reliable > ordered > > messaging between nodes of the cluster. > > > > How does any subsystem relying on CI make sure > > that messages sent by any node in the cluster to > > everybody else is seen in the same order on all > the > > nodes. Isn't this kind of feature neccessary for > SSI > > or any other clients of CI? If so how is that > > accomplished? > > > > > > Thanks, > > Ram Pai > > > > __________________________________________________ > > Do You Yahoo!? > > Get personalized email addresses from Yahoo! Mail > > http://personal.mail.yahoo.com/ > > _______________________________________________ > > ci-linux-devel mailing list > > ci-...@op... > > > http://www.opensource.compaq.com/mailman/listinfo/ci-linux-devel > ===== Ram PaiIBM Corporation,Beaverton OR-...@ya... or pa...@us... __________________________________________________ Do You Yahoo!? Get personalized email addresses from Yahoo! Mail http://personal.mail.yahoo.com/ |
|
From: Brian J. W. <Bri...@co...> - 2001-06-28 02:21:28
|
CI 0.5.5 and Cluster Tools 0.5.5 are now available. As I said a couple
of days ago, the primary change is in how a node joins a cluster.
Previously, it joined before mounting the root and got its information
through LILO. Now it joins when a command is run (sometime after
mounting the root), and it gets its information from a configuration
file.
The source is available on our project summary page:
http://opensource.compaq.com/sourceforge/project/?group_id=22
Hopefully we'll get CVS up soon, so that those interested in the
bleeding edge won't have to wait for a release. We have a few issues
to work through first.
-Brian
|
|
From: Bruce W. <br...@ka...> - 2001-06-27 23:46:56
|
Ram, Thanks for the interest. We do not have specific support for global reliable ordered messaging. The reason is that in all our years of SSI kernel clustering, we have rarely needed it. Algorithms that require talking to all nodes in the cluster do not scale well and we architectually avoid them whenever possible. One exception to that is the Cluster Membership Service. The current master node must step all the nodes through a couple of steps when nodes come up or down. To do that, we tend to use the request/response (also knows as async RPC) feature of ICS. Using the interface to the CLMS, a subsystem can determine the nodes to send a given message to. He then can send them all (no interface in ICS to sendall so this is a call per node), and then wait for all the responses, after which he can go on to the the next phase. Are there specific algorthims that you feel need this kind of capability. If so, maybe we can discuss the need and maybe add this kind of capability. bruce > Hi, > Trying to educate myself on CI. > > Looking at the features published on the web page, > I did not find any support for global reliable ordered > messaging between nodes of the cluster. > > How does any subsystem relying on CI make sure > that messages sent by any node in the cluster to > everybody else is seen in the same order on all the > nodes. Isn't this kind of feature neccessary for SSI > or any other clients of CI? If so how is that > accomplished? > > > Thanks, > Ram Pai > > __________________________________________________ > Do You Yahoo!? > Get personalized email addresses from Yahoo! Mail > http://personal.mail.yahoo.com/ > _______________________________________________ > ci-linux-devel mailing list > ci-...@op... > http://www.opensource.compaq.com/mailman/listinfo/ci-linux-devel |
|
From: s J. <jpn...@ya...> - 2001-06-27 13:22:56
|
Test Mail for checking Mailing list. __________________________________________________ Do You Yahoo!? Get personalized email addresses from Yahoo! Mail http://personal.mail.yahoo.com/ |
|
From: David B. Z. <dav...@co...> - 2001-06-27 04:38:42
|
Ramachandra Pai wrote: > Hi, > Trying to educate myself on CI. > > Looking at the features published on the web page, > I did not find any support for global reliable ordered > messaging between nodes of the cluster. Our ICS communication layer guarantees reliable messages between nodes. Ordering is NOT performed by the communication layer. We have multiple service daemons available to process incoming messages, even if the messages were ordered, we'd have to limit the service daemons in order to guarantee that they are processed in the order they arrived. > > > How does any subsystem relying on CI make sure > that messages sent by any node in the cluster to > everybody else is seen in the same order on all the > nodes. Isn't this kind of feature neccessary for SSI > or any other clients of CI? If so how is that > accomplished? We built full SSI on UNIXWARE without complete ordering between nodes. I worked on a filesystem token (lock) manager which ordered just the token messages between nodes. This mechanism was component specific and did not interact with any other sub-system. The token sub-system ordering took 450 lines of code to perform. Other sub-systems do not perform any ordering. As a matter of fact when we wrote the code, we always assumed that a receiving node may "process" incoming messages in order. Think of an SMP machine which has multiple service daemons handling incoming messages. That means multiple message are even handled simultaneously! -- David B. Zafman | Compaq Computer Corporation mailto:dav...@co... | http://www.compaq.com "Thus spake the master programmer: When you have learned to snatch the error code from the trap frame, it will be time for you to leave." |
|
From: David B. Z. <dav...@co...> - 2001-06-27 03:57:39
|
"Brian J. Watson" wrote: > How this would work for an SSI cluster is that the node would mount a > local ramdisk and start a modified linuxrc as process 1. In addition > to its usual stuff, linuxrc would run our command to join the cluster, > then it would pivot_root() the clusterwide shared root into place. > Finally, it would exec init if it's the first node in the cluster, or > exit otherwise. (Please correct me if I'm wrong, Dave) We make the first process pid 2 (intead of 1 as in base LINUX), so that only if init needs to get spawned do we specifically create pid 1 and exec init, otherwise, if an init is NOT needed (dependent node), the pid 2 process just sleeps forever. We don't have pid 2 exit because all the system daemons are its children and we don't want them inherited by process 1. -- David B. Zafman | Compaq Computer Corporation mailto:dav...@co... | http://www.compaq.com "Thus spake the master programmer: When you have learned to snatch the error code from the trap frame, it will be time for you to leave." |
1 message has been excluded from this view by a project administrator.
