Hello, Did anyone do some work on improving the samba read/write performance? We've got a board attached with an IDE hadisk, running samba 2.0.7 on uClinux(NOMMU) OS, the IDE throughput: 10.5Mbps,ethernet speed: 16Mbps,Samba throughput(copy file from WINDOWS PC to host board ): 600KByte/s,CPU working at 120MHz, system SDRAM 16Mbyte. Could anyone help to figure out where the bottleneck is and how to tune it? Thanks a lot! Happy hacking... hn _________________________________________________________ Do You Yahoo!? ÑÅ»¢Í¨ÍøÂçKTV, ËæʱËæµØÃâ·Ñ¿¨ÀOK~~ http://rd.yahoo.com/mail_cn/tag/?http://cn.messenger.yahoo.com//chat/index.html
David Collier-Brown -- Customer Engineering
2003-Apr-16 12:52 UTC
[Samba] Re: samba performance on uclinux embedded system
[starting to get off-topic, moved to samba@lists.samba.org] I reccomend RAID striped across six drives, so that there is (nearly) always a physical head available to read or write the data. This tends to hide the normal elevator-like scan algorithm used by Unix from the program, and delivers higher throughput. --dave hong joy wrote:> Dave, > We write a program to open a file and then write file > to test the IDE and got the IDE throughput 10.5M bps. > > What is 'a 6-wide stripe' ? > > We try to do zero-copy for file-open-write, just > something like zero-copy in linux tcpip stack. Any > advice? > > Thanks a lot! Happy hacking! > > Regards, > Joy > > --- David Collier-Brown -- Customer Engineering > <David.Collier-Brown@Sun.COM> µÄÕýÎÄ£º> > >>hong joy wrote: >> >>>Hello, >>>Did anyone do some work on improving the samba >>>read/write performance? >>>We've got a board attached with an IDE hadisk, >> >>running >> >>>samba 2.0.7 on uClinux(NOMMU) OS, the IDE >> >>throughput: >> >>>10.5Mbps,ethernet speed: 16Mbps,Samba >> >>throughput(copy >> >>>file from WINDOWS PC to host board ): >> >>600KByte/s,CPU >> >>>working at 120MHz, system SDRAM 16Mbyte. >>>Could anyone help to figure out where the >> >>bottleneck >> >>>is and how to tune it? >>>Thanks a lot! >>>Happy hacking... >> >> Even assuming a 10,000-rpm disk, you'll bottleneck >>on >>disk I/O: a single disk head with the normal Unix >>elevator algorithm (ext2 or 3 fs) will peak out at >>about 800 KB/S, because the total sequential-read >>throughput can't be acheived, courtesy of all the >>seeking about. I'd reccomend about a 6-wide stripe >>or better. >> >>--dave >> >> >> >> >>>hn >>> >>> >>> >> > _________________________________________________________ > >>>Do You Yahoo!? >>>ÑÅ»¢Í¨ÍøÂçKTV, ËæʱËæµØÃâ·Ñ¿¨ÀOK~~ >>> >> > http://rd.yahoo.com/mail_cn/tag/?http://cn.messenger.yahoo.com//chat/index.html > >> >>-- >>David Collier-Brown, | Always do right. >>This will gratify >>Sun Microsystems DCMO | some people and >>astonish the rest. >>Toronto, Ontario | >>(905) 415-2849 or x52849 | >>davecb@canada.sun.com >> >> > > > _________________________________________________________ > Do You Yahoo!? > ÑÅ»¢ÑûÄã²ÎÓëCCTV-2¶Ô»°½ÚÄ¿µ÷²é£ºÒƶ¯¡¢ÁªÍ¨¡¢Áéͨ£¬ÒªËµ°®Äã²»ÈÝÒ× > http://rd.yahoo.com/mail_cn/tag/?http://cn.surveys.yahoo.com/cctv_wireless_study >-- David Collier-Brown, | Always do right. This will gratify Sun Microsystems DCMO | some people and astonish the rest. Toronto, Ontario | (905) 415-2849 or x52849 | davecb@canada.sun.com