I have news! I had too much time ;) so I recompiled partclone for profiling with gprof.
The terrifying result: partclone wastes almost 90% of the time with logging, calculating the speed and updating the gui. (the crc checks need as well some time). (for example it does such bad things like trying to refresh the gui for EACH block, which is toooo much ;) ) I'll give you the complete gprof results later. I'am atm buzy with testing. I removed for testing just for fun all functions which needed much time as per gprof and OH WONDER, the speed growed from 5-6 MB/s to 20-35 MB/s. On 12.04.2010 03:39, Steven Shiau wrote: > > > On 2010/4/9 下午 03:50, Lukas Grässlin wrote: >> On 31.03.2010 11:10, Steven Shiau wrote: >>> Right now I do not have real machines which I can test. Therefore I can >>> not give you the numbers. >>> If anyone on this forum has such numbers to share, please share that. >>> >>> BTW, there is a performance improvement in partclone 0.2.8, and it's now >>> included in clonezilla live 20100330-karmic. Could you please give it a >>> try? To see if any big difference. >>> Please let us know the results if you try that. >> >> So, I tried the 20100330 clonezilla ISO and the results with the speed >> are the same. I think the main reason for that bad speed is, that on the >> client side (the vm, where the physical machine is migrated to) the >> partclone.restore process procudes almost 100% CPU load. (It's s a vm >> with two cores, but it only uses one). >> I think that is the main bottleneck. > What's the memory size you have on the vm? > > Steven. >> >>> >>> Regards, >>> Steven. >>> >>> On 2010/3/29 下午 11:31, Lukas Grässlin wrote: >>>> Both SATA Disks, the destination is a virtual machine but I did some >>>> IO-Performance tests with dd on the virtual machine. It is definitely >>>> able to write and read with more than 20MB/s. (I did dd if=/dev/sda >>>> of=/dev/zero bs=100M count=10 etc.) >>>> >>>> The network can't really limit the speed, so I don't know what is >>>> could be. >>>> >>>> What's your experience with the speed? Is it faster? >>>> >>>> I'll do some tests on my own with dd and netcat or so. >>>> >>>> ((sorry, forgot to click the reply-all button ;) )) >>>> >>>> >>>> On 29.03.2010 16:26, Steven Shiau wrote: >>>>> How about the speed when you save the image? >>>>> What's the disk types in the source and destination machines? SATA? >>>>> PATA? USB? Or? >>>>> >>>>> Steven. >>>>> >>>>> Lukas Grässlin wrote: >>>>>> Hi, >>>>>> >>>>>> whats's you experience with the onthefly migration (that partclone >>>>>> over >>>>>> netcat thing) especially perfomance? >>>>>> >>>>>> I never get more than ~300MB/min (=~ 5MB/s) in a Gigabit network >>>>>> which >>>>>> is very dissappoiting. I already tried it without compression etc >>>>>> but I >>>>>> didn't get more speed. (Further the machines are fast enough to do it >>>>>> faster than 5MB/s with compression). >>>>>> >>>>>> Is that a partclone issue? Have you any ideas? >>>>>> >>>>>> Regards, >>>>>> Lukas >>>>>> >>>>>> >>>>> >>>> >>>> >>> >> >> > -- Lukas Grässlin Collax GmbH . Basler Str. 115a . 79115 Freiburg . Germany p: +49 (0) 89-990 157-23 Collax - Simply Linux. Geschäftsführer: Boris Nalbach AG München HRB 158898 * Ust.-IdNr: DE 814464942 ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Clonezilla-live mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/clonezilla-live
