ssic-linux-users Mailing List for OpenSSI Clusters for Linux (Page 10)
Brought to you by:
brucewalker,
rogertsang
Menu
▾
▴
ssic-linux-users — User discussion
You can subscribe to this list here.
| 2003 |
Jan
(17) |
Feb
(23) |
Mar
(32) |
Apr
(48) |
May
(51) |
Jun
(23) |
Jul
(39) |
Aug
(47) |
Sep
(107) |
Oct
(112) |
Nov
(112) |
Dec
(70) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2004 |
Jan
(155) |
Feb
(283) |
Mar
(200) |
Apr
(107) |
May
(73) |
Jun
(171) |
Jul
(127) |
Aug
(119) |
Sep
(91) |
Oct
(116) |
Nov
(175) |
Dec
(143) |
| 2005 |
Jan
(168) |
Feb
(237) |
Mar
(222) |
Apr
(183) |
May
(111) |
Jun
(153) |
Jul
(123) |
Aug
(43) |
Sep
(95) |
Oct
(179) |
Nov
(95) |
Dec
(119) |
| 2006 |
Jan
(39) |
Feb
(33) |
Mar
(133) |
Apr
(69) |
May
(22) |
Jun
(40) |
Jul
(33) |
Aug
(32) |
Sep
(34) |
Oct
(10) |
Nov
(8) |
Dec
(18) |
| 2007 |
Jan
(14) |
Feb
(3) |
Mar
(13) |
Apr
(16) |
May
(15) |
Jun
(8) |
Jul
(20) |
Aug
(25) |
Sep
(17) |
Oct
(10) |
Nov
(8) |
Dec
(13) |
| 2008 |
Jan
(7) |
Feb
|
Mar
(1) |
Apr
(6) |
May
(15) |
Jun
(22) |
Jul
(22) |
Aug
(5) |
Sep
(5) |
Oct
(17) |
Nov
(3) |
Dec
(1) |
| 2009 |
Jan
(2) |
Feb
|
Mar
(29) |
Apr
(78) |
May
(17) |
Jun
(3) |
Jul
|
Aug
|
Sep
(1) |
Oct
(21) |
Nov
(1) |
Dec
(4) |
| 2010 |
Jan
(1) |
Feb
(5) |
Mar
|
Apr
(5) |
May
(7) |
Jun
(14) |
Jul
(5) |
Aug
(72) |
Sep
(25) |
Oct
(5) |
Nov
(14) |
Dec
(12) |
| 2011 |
Jan
(9) |
Feb
|
Mar
|
Apr
(3) |
May
(3) |
Jun
(2) |
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2012 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
(10) |
Aug
(18) |
Sep
(2) |
Oct
(1) |
Nov
|
Dec
|
| 2013 |
Jan
(1) |
Feb
(3) |
Mar
|
Apr
(2) |
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
(1) |
Dec
|
| 2014 |
Jan
|
Feb
|
Mar
|
Apr
(2) |
May
(1) |
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
| 2019 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
(1) |
Jul
|
Aug
|
Sep
|
Oct
|
Nov
|
Dec
|
|
From: Christopher G. S. I. <cg...@ld...> - 2009-12-22 13:26:21
|
----- "James T Studebaker" <jts...@ji...> wrote: > Has anyone been able to set up VM software to run on an openssi > cluster to make all the resources of the cluster available to the VM > software? Treating an OpenSSI cluster as a big NUMA machine shouldn't be _that_ difficult, but certainly not a walk in the park. Getting OpenSSI smooshed into any modern kernel would still be necessary, though. A stable target like RHEL/CentOS 5 is a good idea, but who has the time these days? :P -- Christopher G. Stach II http://ldsys.net/~cgs/ |
|
From: Robert W. <no...@ro...> - 2009-12-22 12:43:07
|
On Tue, Dec 22, 2009 at 04:59:16AM -0500, James T Studebaker wrote: > Has anyone been able to set up VM software to run on an openssi cluster to > make all the resources of the cluster available to the VM software? You mean, some VM software running in openssi as host system, where the guests can use all the ressources of the whole cluster? I never tried that, but I think, the main problem would be, if you want to run a guest as a virtual multiprocessor system, and run the processes inside the guest on different hosts of the cluster. I guess, that this is not possible at the moment (because threads have to run on one clusternode). On the other side, running one virtual machine on one clusternode shouldn't be a big problem. I hope. Nevertheless, a good idea to try that. Robert |
|
From: James T S. <jts...@ji...> - 2009-12-22 09:59:28
|
Has anyone been able to set up VM software to run on an openssi cluster to make all the resources of the cluster available to the VM software? |
|
From: Daniel B. <dan...@gm...> - 2009-11-20 18:39:41
|
Hi, I was looking around the wiki for OpenSSI and didn't find any information about somebody installing OpenSSI on CentOS 5.4 I downloaded openssi-fc3-1.9.2.i686.tar.gz and when I run the install script I get a bunch of errors about missing some packages that I already have on a higher version. Any of you has some info on how to get around this? This is what I get [root@centos openssi-fc3-1.9.2.i686]# ./install This version of OpenSSI 1.9 is intended for a system installed with Fedora Core 3. You should run the install script on a fresh Fedora Core 3 installation, or download the correct version of OpenSSI 1.9 for your distribution. Versions for Debian and SuSE are also available. Would you like to install this version of OpenSSI anyway (y/n) [n]: y Checking for base packages... checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking Checking for OpenSSI packages... checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking checking warning: RPMS/base/dhcp-3.0.1-44_FC3.i386.rpm: Header V3 DSA signature: NOKEY, key ID 4f2a6fd2 error: Failed dependencies: /usr/lib/python2.3 is needed by libxml2-python-2.6.16-3.i386 beecrypt >= 3.0.1 is needed by rpm-4.3.2-21.i386 libbeecrypt.so.6 is needed by rpm-4.3.2-21.i386 librpm-4.3.so is needed by rpm-4.3.2-21.i386 librpmbuild-4.3.so is needed by rpm-4.3.2-21.i386 librpmdb-4.3.so is needed by rpm-4.3.2-21.i386 librpmio-4.3.so is needed by rpm-4.3.2-21.i386 popt = 1.9.1 is needed by rpm-4.3.2-21.i386 initscripts >= 8.11.1-1 is needed by kernel-2.6.18-164.el5.i686 initscripts < 8.40-1 conflicts with kudzu-1.2.57.1.21-1.el5.centos.i386 nfs-utils < 1.0.7-12 conflicts with kernel-2.6.18-164.el5.i686 libnewt.so.0.51 is needed by ntsysv-1.3.11.1-1_ssi_1devel.i386 db4 = 4.2.52 is needed by pam-0.77-65_ssi_1devel.i386 pam < 0.78 conflicts with setuptool-1.19.2-1.el5.centos.i386 procps < 3.2.5-6.3 conflicts with kernel-2.6.18-164.el5.i686 hotplug is needed by udev-039-10.FC3.7_ssi_1devel.i386 udev < 063-6 conflicts with kernel-2.6.18-164.el5.i686 libxml2 >= 2.6.25 is needed by (installed) libxslt-1.1.17-2.el5_2.2.i386 rpm = 4.4.2.3-18.el5 is needed by (installed) rpm-libs-4.4.2.3-18.el5.i386 rpm = 4.4.2.3-18.el5 is needed by (installed) rpm-python-4.4.2.3-18.el5.i386 rpm >= 0:4.4.2 is needed by (installed) yum-3.2.22-20.el5.centos.noarch rpm = 4.4.2.3-18.el5 is needed by (installed) rpm-build-4.4.2.3-18.el5.i386 chkconfig >= 1.3.26 is needed by (installed) dbus-glib-0.73-8.el5.i386 chkconfig >= 1.3.26 is needed by (installed) dbus-1.1.2-12.el5.i386 initscripts >= 8.11.1-1 is needed by (installed) kernel-2.6.18-164.el5.i686 initscripts >= 8.04-1 is needed by (installed) hal-0.5.8.1-52.el5.i386 mkinitrd >= 4.2.21-1 is needed by (installed) kernel-2.6.18-164.el5.i686 libpam.so.0(LIBPAM_1.0) is needed by (installed) coreutils-5.97-23.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) ppp-2.4.4-2.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) libuser-0.54.7-2.el5.5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) passwd-0.73-1.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) usermode-1.88-3.el5.2.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) at-3.1.8-82.fc6.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) vixie-cron-4.1-76.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) cyrus-sasl-2.1.22-5.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) krb5-workstation-1.6.1-36.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) nss_ldap-253-21.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) pam_pkcs11-0.5.3-23.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) pam_ccreds-3-5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) pam_krb5-2.2.14-10.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) policycoreutils-1.33.12-14.6.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) cups-1.3.7-11.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) openssh-server-4.3p2-36.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) cvs-1.11.22-7.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) screen-4.0.3-1.el5.i386 libpam.so.0(LIBPAM_1.0) is needed by (installed) squid-2.6.STABLE21-3.el5.i386 libpam.so.0(LIBPAM_EXTENSION_1.0) is needed by (installed) pam_pkcs11-0.5.3-23.i386 libpam_misc.so.0(LIBPAM_MISC_1.0) is needed by (installed) coreutils-5.97-23.el5.i386 libpam_misc.so.0(LIBPAM_MISC_1.0) is needed by (installed) libuser-0.54.7-2.el5.5.i386 libpam_misc.so.0(LIBPAM_MISC_1.0) is needed by (installed) passwd-0.73-1.i386 libpam_misc.so.0(LIBPAM_MISC_1.0) is needed by (installed) usermode-1.88-3.el5.2.i386 libpam_misc.so.0(LIBPAM_MISC_1.0) is needed by (installed) policycoreutils-1.33.12-14.6.el5.i386 pam >= 0.78 is needed by (installed) passwd-0.73-1.i386 pam >= 0.99.6.2-3.27 is needed by (installed) authconfig-5.3.21-6.el5.i386 pam >= 0.77-66.8 is needed by (installed) vixie-cron-4.1-76.el5.i386 /lib/security/pam_loginuid.so is needed by (installed) openssh-server-4.3p2-36.el5.i386 udev >= 062 is needed by (installed) pcmciautils-014-5.i386 udev >= 039-10.14.EL4 is needed by (installed) isdn4k-utils-3.2-56.el5.i386 udev >= 089-1 is needed by (installed) hal-0.5.8.1-52.el5.i386 Failed to install packages [root@centos openssi-fc3-1.9.2.i686]# Any help will be greatly appreciated. Best regards, -- Dan Bahena http://twitter.com/informatux http://xti.informatux.net |
|
From: John H. <jo...@Ca...> - 2009-10-20 14:48:37
|
John Hughes wrote:
> Here's my /etc/xen/scripts/openssi-bridge script to set up two bridges:
>
> #!/bin/sh
> # network-xen-custom
> # Exit if anything goes wrong
> set -e
>
> interfaces="calvaedi openssi"
>
^ obviously should be
interfaces="eth0 eth1"
> dir=`dirname $0`
> script=$dir/network-bridge
>
> for i in ${interfaces}
> do
> $script $1 netdev=$i
> done
>
|
|
From: John H. <jo...@Ca...> - 2009-10-20 10:53:05
|
Anyone wanting to play with OpenSSI but doesn't want to dedicate a
machine or machines, just a little notice that it seesm to work OK under
Xen HVM (hardware virtual machine, on modern hardware that has HVM
instructions, e.g. Intel Core or recent AMD's). To run in PVM mode the
old OpenSSI/Xen patches would need resurrection.
Example Xen config:
kernel = '/usr/lib/xen-3.2-1/boot/hvmloader'
builder = 'hvm'
memory = '256'
vcpus = 2
device_model='/usr/lib/xen-3.2-1/bin/qemu-dm'
# Disks
disk = [
'file:/home/xen/node1.img,hda,w',
'phy:/dev/some_vg/openssi.img,hdb,w' ]
# Hostname
name = 'node1'
# Networking, one RTL 8139+ for external network and one for
# interconnect.
# Need specialised bridge script.
vif = ['type=ioemu, bridge=eth0, mac=00:16:3E:52:c1:9B',
'type=ioemu, bridge=eth1, mac=00:16:3E:52:d1:9B' ]
# Behaviour
boot='c'
# console
sdl=0
vnc=1
vnclisten='0.0.0.0'
Here's my /etc/xen/scripts/openssi-bridge script to set up two bridges:
#!/bin/sh
# network-xen-custom
# Exit if anything goes wrong
set -e
interfaces="calvaedi openssi"
dir=`dirname $0`
script=$dir/network-bridge
for i in ${interfaces}
do
$script $1 netdev=$i
done
|
|
From: John H. <jo...@Ca...> - 2009-10-13 09:41:16
|
Advice Pro wrote: > What I meant by What does this suggest, is If I should create multiple > partitions to dynamically re-configure partitions on an ordinary OpenSSI > cluster. If it's possible, why would I want to dynamically reconfigure > partitions. > > OpenSSI doesn't do "partitions". What you get with OpenSSI: You take a bunch of PC, with various hardware bits (serial ports, USB devices, video cards, disks and so on), you connect 'em all together with a gigabit ethernet LAN, you run an OpenSSI kernel on 'em all and bingo! You have one big machine. Any process runnning on any "node" (those PCs) can access the hardware on any other nodes. Processes can be transparently moved from node to node and can start children on any node they want. If you configure things right you can arrange that the cluster keeps running even if a node crashes. That's it. (Anything I've forgotten?) |
|
From: Advice P. <LA...@gm...> - 2009-10-12 14:32:27
|
What I meant by What does this suggest, is If I should create multiple partitions to dynamically re-configure partitions on an ordinary OpenSSI cluster. If it's possible, why would I want to dynamically reconfigure partitions. > I didn't understand this thread though it should prove useful: > > http://www.nabble.com/Performance-Tests-td2013120.html#a2013120 > > I want to know ways are there to monitor & test performance of OpenSSI, > especially load balancing. run something like PVM pov ray, parallel john the ripper etc -- View this message in context: http://www.nabble.com/Performance-and-Performance-Testing--tp25846714p25856820.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: Advice P. <LA...@gm...> - 2009-10-12 14:07:30
|
I understand why OpenSSI might not cluster for games, but how about video editing or home design? -- View this message in context: http://www.nabble.com/Does-OpenSSI-combine-video-from-motherboard-chipsets-and-video-cards--tp25842806p25856451.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: Mulyadi S. <mul...@gm...> - 2009-10-12 03:01:20
|
On Sun, Oct 11, 2009 at 10:19 PM, Advice Pro <LA...@gm...> wrote: > > The only information that I have swapped spaces is > http://openssi.org/cgi-bin/view?page=docs2/1.2/Introduction-to-SSI > VII: Swap > Swap space is not particularly SSI, with each node providing its own > swap space. Swap space designation is done in the common /etc/fstab. > To determine which nodes are using which swap, you can execute an > onall swapon -s. Probably it's a pool of active swap partitions/files configured on each node of openSSI cluster. IMO process are still swapping in/out to its home node, so this "swap space" might be needed to define the relationship -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com |
|
From: Mulyadi S. <mul...@gm...> - 2009-10-12 02:55:38
|
On Mon, Oct 12, 2009 at 2:26 AM, Advice Pro <LA...@gm...> wrote: > > What does this suggest: > > http://www.nabble.com/XEN%2BOpenSSI%2C-big-performace-overhead--td5643723.html#a5643723 > Aggregating as a big system and partitioning will create big > performance overhead and is inefficient. Instead you can create > multiple partitions with SSI interface itself if you work on > hot-plug of systems which allows you to dynamically re-configure your > partitions" > > That I should simply create multiple partions, or what does it mean with SSI > interface?And to what is hot plug up systems which allows you to dynamically > re-configure your partions? Looks the original post talks about load balancing Xen virtual machine on top of openSSI. If it is, better use the Xen capability to do VM migration between Xen instance. Almost every known virtualization technologies have this capability. > I didn't understand this thread though it should prove useful: > > http://www.nabble.com/Performance-Tests-td2013120.html#a2013120 > > I want to know ways are there to monitor & test performance of OpenSSI, > especially load balancing. run something like PVM pov ray, parallel john the ripper etc -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com |
|
From: Mulyadi S. <mul...@gm...> - 2009-10-12 02:48:08
|
On Mon, Oct 12, 2009 at 5:35 AM, Advice Pro <LA...@gm...> wrote: > > http://openssi.org/cgi-bin/view?page=docs2/1.2/README-mosixll This page on > OpenSSI.org says: > > Once the machines are up loadleveling can be started using the > "loadlevel -a on" command to turn loadleveling on for all nodes > in the cluster. OK, and your specific question is...? -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com |
|
From: Mulyadi S. <mul...@gm...> - 2009-10-12 02:47:12
|
On Sun, Oct 11, 2009 at 7:02 PM, Advice Pro <LA...@gm...> wrote: > > Some applications I like to use this for is gaming & video editing and > playing. > > Based on the answer, I'd assume the same for Raid cards and motherboard > chipsets. AFAIK, games won't take any advantage of current clustering technology even it is SSI type of clustering. So, unless you get a specific port of a game to harness clustering, you won't see things like combined power of video processing. As alternative, read about CUDA and buy games that specifically use this specific NVIDIA innovation. -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com |
|
From: Advice P. <LA...@gm...> - 2009-10-11 22:35:15
|
http://openssi.org/cgi-bin/view?page=docs2/1.2/README-mosixll This page on OpenSSI.org says: Once the machines are up loadleveling can be started using the "loadlevel -a on" command to turn loadleveling on for all nodes in the cluster. -- View this message in context: http://www.nabble.com/What-do-I-really-need-to-know-for-load-balancing-in-1.2--tp25848302p25848302.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: Advice P. <LA...@gm...> - 2009-10-11 19:25:46
|
What does this suggest: http://www.nabble.com/XEN%2BOpenSSI%2C-big-performace-overhead--td5643723.html#a5643723 Aggregating as a big system and partitioning will create big performance overhead and is inefficient. Instead you can create multiple partitions with SSI interface itself if you work on hot-plug of systems which allows you to dynamically re-configure your partitions" That I should simply create multiple partions, or what does it mean with SSI interface?And to what is hot plug up systems which allows you to dynamically re-configure your partions? I comfortably didn't understand this thread though it should prove useful: http://www.nabble.com/Performance-Tests-td2013120.html#a2013120 I want to know ways are there to monitor & test performance of OpenSSI, especially load balancing. -- View this message in context: http://www.nabble.com/Performance-and-Performance-Testing--tp25846714p25846714.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: Advice P. <LA...@gm...> - 2009-10-11 17:41:05
|
I think what I need is two PCs, one for the master node and another for failover, etc. And of course I need a network switch. Won't any Linux compatible hardware work with OpenSSI? The reason I asked is because I read: http://www.nabble.com/Installing-OpenSSI-on-a-SATA-hdd-td8120175.html#a8120175 -- View this message in context: http://www.nabble.com/What-is-the-exact-hardware-that-I-need-for-an-OpenSSI-cluster--tp25845731p25845731.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: Advice P. <LA...@gm...> - 2009-10-11 15:19:47
|
The only information that I have swapped spaces is http://openssi.org/cgi-bin/view?page=docs2/1.2/Introduction-to-SSI VII: Swap Swap space is not particularly SSI, with each node providing its own swap space. Swap space designation is done in the common /etc/fstab. To determine which nodes are using which swap, you can execute an onall swapon -s. I've read on the forum (something like) swap space is on a partion. So what exactly is swap space and where would it be in the node? -- View this message in context: http://www.nabble.com/I-don%27t-completely-understand-what-is-swap-space%21-tp25844279p25844279.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: Advice P. <LA...@gm...> - 2009-10-11 12:02:27
|
Some applications I like to use this for is gaming & video editing and playing. Based on the answer, I'd assume the same for Raid cards and motherboard chipsets. -- View this message in context: http://www.nabble.com/Does-OpenSSI-combine-video-from-motherboard-chipsets-and-video-cards--tp25842806p25842806.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: pk <pe...@mi...> - 2009-10-08 18:00:53
|
John Hughes wrote: > Shared memory between OpenSSI nodes "works" but you don't want to be > using it in a multiple-writer situation like most threaded programs. Thanks for the explanation. I really appreciate it! I only wish I had the skills to contribute... Best regards Peter K |
|
From: John H. <jo...@Ca...> - 2009-10-08 10:54:06
|
pk wrote:
> John Hughes wrote:
>
>
>> Multi-threaded apps run all threads on the same node. (This is because
>> sharing writable memory between nodes is pretty inefficient). If you
>> write your app as "multi-process" rather than "multi-thread",
>> communicating using pipes, sockets or queues instead of shared memory
>> then you can distribute your app across multiple nodes.
>>
>
> Sorry for "butting" in...
>
> I'm just curious about what happened to the thread migration plans or
> perhaps I remember that from another project? OpenMosix maybe? As I
> understand it thread migration requires shared memory. Of course this
> will introduce huge differences depending on the connection between
> nodes but can't a NUMA model, perhaps with rDMA, handle such things?
> Again, I'm just curious about the subject and don't know much about it...
>
In most OpenSSI clusters the connection between nodes is *many* times
slower than the cpu-memory connection, historically it has been 10mbit
ethernet (LOCUS), Compaq Servernet or 100mbit ethernet (UnixWare
clusters), 1000mbit ethernet (OpenSSI).
Shared memory between OpenSSI nodes "works" but you don't want to be
using it in a multiple-writer situation like most threaded programs.
How shared memory works -
If a process writes a shared memory page then either:
1. it is currently owned by (writable by) the node running the
process, nothing special happens, the process writes the page as usual
2. it has been written or read by processes on some other node or
node, then the process gets a page fault and the cluster system
makes an inter-node call to the other nodes, "stealing" the page,
it is marked unreadable/unwritable on the other nodes and writable
on this node.
If a process reads a shared memory page then either:
1. It is currently readable by the node running the process. Nothing
special happens.
2. Some other node has a readable copy - an inter-node call is made
to get a copy of the page.
3. It is currently writable by some other node - an inter-node call
is made to get a copy of the page and mark it unwritable.
So any page can have one writer, or multiple readers. Multiple read
access is fast, but once someone starts writing you get page faults and
inter-node IPC.
In other words OpenSSI has an extremely NUMA memory model.
In this situation allowing different threads of a process, which by
definition share memory, to run on different nodes doesn't seem reasonable.
What we do allow (or should allow) is for all threads of a process to
migrate to some other node in one go.
|
|
From: pk <pe...@mi...> - 2009-10-07 19:49:21
|
John Hughes wrote: > Multi-threaded apps run all threads on the same node. (This is because > sharing writable memory between nodes is pretty inefficient). If you > write your app as "multi-process" rather than "multi-thread", > communicating using pipes, sockets or queues instead of shared memory > then you can distribute your app across multiple nodes. Sorry for "butting" in... I'm just curious about what happened to the thread migration plans or perhaps I remember that from another project? OpenMosix maybe? As I understand it thread migration requires shared memory. Of course this will introduce huge differences depending on the connection between nodes but can't a NUMA model, perhaps with rDMA, handle such things? Again, I'm just curious about the subject and don't know much about it... Hope I'm making some sense... Best regards Peter K |
|
From: John H. <jo...@Ca...> - 2009-10-07 14:40:33
|
Advice Pro wrote: > A > http://www.nabble.com/Does-OpenSSI-combine-CPU-cores-for-multithreaded-applications--td23124003.html > question that I had asked was answered that "All threads of a process must > run on the same node at the moment". I think this means that only one > multicore CPU is ideal per node. > ??? You need at least one cpu per node, of course. :-) You can use many as you feel like. Multi-threaded apps run all threads on the same node. (This is because sharing writable memory between nodes is pretty inefficient). If you write your app as "multi-process" rather than "multi-thread", communicating using pipes, sockets or queues instead of shared memory then you can distribute your app across multiple nodes. > Where on OpenSSI.org does it state such things > |
|
From: Advice P. <LA...@gm...> - 2009-10-06 18:17:34
|
A http://www.nabble.com/Does-OpenSSI-combine-CPU-cores-for-multithreaded-applications--td23124003.html question that I had asked was answered that "All threads of a process must run on the same node at the moment". I think this means that only one multicore CPU is ideal per node. Where on OpenSSI.org does it state such things -- View this message in context: http://www.nabble.com/How-much-CPU-power---memory-per-node--tp25773695p25773695.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
|
From: John H. <jo...@Ca...> - 2009-10-06 15:54:47
|
Advice Pro wrote: > Or is it just academic & scientific software only? > I'm running gnome & openoffice & firefox and stuff like that. |
|
From: Advice P. <LA...@gm...> - 2009-10-05 21:20:40
|
Or is it just academic & scientific software only? -- View this message in context: http://www.nabble.com/Can-OpenSSI-run-just-any-Linux-compatible-software--tp25757020p25757020.html Sent from the ssic-linux-users mailing list archive at Nabble.com. |
22 messages has been excluded from this view by a project administrator.
