From: thunderlord <thu...@ra...> - 2005-09-18 10:29:59
|
Hi again! I understand that using a global cluster filesystem would be a great=20 choice, but for my small cluster of 2, maybe 3 nodes I think it is not=20 necessary. I just want to speed up data transfer between stub process on=20 home node and worker process on another node. Currently it looks like=20 some bandwidth limit is set for each migrated process, limiting=20 bandwidth to 800-900 Kbit/s, so the only way to make my 935Mhz Duron to=20 work at ~99% CPU usage is to migrate enough processes, then overall=20 badwidth grows in arithmetic progression, but this is not always acceptab= le. *So my question is: does anybody know where in kernel code could be such=20 a function/piece of code that limits data transfer rates between=20 stub(deputy) process (which fetches data from harddisk on home node) and=20 worker process on another node?* I analized changes that openmosix patch make to vanilla kernel code and=20 found several possible places, but to test them all can take a lot of=20 time, because i am not so good in kernel development, and do not know=20 openmosix kernel code good enough. Thanks in advance. Evan Hisey wrote: >Thunderlord- > That would seem to be right. oM is ideal crunching numbers not >accessing harddrives across a network. I think you may also be >confussed about how oM works. It moves the whole proccess to teh new >host, leaving only a stub behind on the lauching node. It does not >'spawn' new versions of the application to move to new nodes. Fora >while oM used DFSA and oMFS to solve the data transfer issue but it >became to ugly to maintain. Current recommendation is to look at >setting up a global cluster filesystem(not NFS). > >Evan > >On 9/15/05, thunderlord <thu...@ra...> wrote: > =20 > >>Hi all! >> >>I am using 2.6.11.12-om and noticed that any migrated process doesn=B4t >>use all bandwidth avaliable for interprocess communication between >>deputy and master processes - maximum is 800-900 Kbit/s for each >>migrated process. So, any program that uses significant amounts of data >>to operate consumes only 3-30% of CPU time (tested on >>lame,ffmpeg,bzip2). I use same kernel on both nodes. I=B4ve tried to >>recompile kernel with different configurations, but no results. Is this >>bug known? How can i fix it? >> >>My configuration: 2 nodes: >>1 - master - 1800 Duron, SDRAM 512Mb >>2 - slave - 935 Duron, SDRAM 256Mb >>Network - Ethernet 100BaseT. >> >> >> >>------------------------------------------------------- >>SF.Net email is sponsored by: >>Tame your development challenges with Apache's Geronimo App Server. >>Download it for free - -and be entered to win a 42" plasma tv or your v= ery >>own Sony(tm)PSP. Click here to play: http://sourceforge.net/geronimo.p= hp >>_______________________________________________ >>openMosix-devel mailing list >>ope...@li... >>https://lists.sourceforge.net/lists/listinfo/openmosix-devel >> >> =20 >> > > =20 > |