David Rees schrieb: > On Tue, Feb 26, 2008 at 4:39 PM, David Rees <[EMAIL PROTECTED]> wrote: >> So there you go. IMO, unless you are willing to overhaul your storage >> system or slightly increase the risk of data corruption (IMO, >> data=writeback instead of the default data=ordered should be a large >> gain for you and is very safe), you are going to continue to fall >> further behind in your nightly cleanup runs. > > I forgot to mention, this link may be informative: > > http://wiki.centos.org/HowTos/Disk_Optimization > > But I think it covers most of the topics in this thread already.
I guess theory and testing that theory on your own system are two different things. Only now I really tested different IO schedulers, to get some numbers. I changed the scheduler on the iSCSI target, as it is the device where the writes take place. After a short period of testing, it seems to me that these are the best schedulers (1 being better than 2, first item seemed only slightly better than the second, i.e. anticipatory was slightly better than noop): 1. anticipatory, or noop 2. deadline, or cfq Note: these results may not be very scientific as the conditions were not constant all the time (BackupPC fetching data from hosts). 1. The Anticipatory elevator introduces a controlled delay before dispatching the I/O to attempt to aggregate and/or re-order requests improving locality and reducing disk seek operations. Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 158,58 2846,31 0,00 28520 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 168,03 1572,03 0,00 15736 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 144,47 1207,11 15554,15 12216 157408 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 85,01 600,20 24815,98 6008 248408 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 119,28 279,72 1537,66 2800 15392 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 164,47 1370,06 598,00 13728 5992 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 151,55 1778,22 0,00 17800 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 189,51 1854,15 0,00 18560 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 118,36 1386,77 19005,73 14048 192528 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 164,84 1914,89 2836,36 19168 28392 2. The NOOP scheduler is a simple FIFO queue and uses the minimal amount of CPU/instructions per I/O to accomplish the basic merging and sorting functionality to complete the I/O. Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 173,63 2318,48 0,00 23208 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 130,27 801,60 16715,28 8024 167320 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 160,04 1774,23 91,91 17760 920 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 117,18 1251,55 0,00 12528 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 166,43 1904,50 0,00 19064 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 168,36 1821,16 0,00 18248 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 187,11 2106,69 0,00 21088 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 133,80 1248,80 13886,40 12488 138864 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 128,10 980,80 529,60 9808 5296 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 178,11 2333,93 125,77 23456 1264 3. The Deadline elevator uses a deadline algorithm to minimize I/O latency for a given I/O request. Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 216,73 2408,76 0,00 24184 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 132,30 1426,40 2812,00 14264 28120 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 205,28 3076,37 0,00 30856 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 190,81 3359,84 0,00 33632 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 185,03 2435,93 0,00 24408 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 161,94 1783,02 0,00 17848 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 186,70 1896,00 0,00 18960 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 117,47 540,52 4150,10 5416 41584 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 115,20 779,20 1395,20 7792 13952 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 123,68 387,61 1509,69 3880 15112 4. CFQ maintains a scalable per-process I/O queue and attempts to distribute the available I/O bandwidth equally among all I/O requests. CFQ is well suited for mid-to-large multi-processor systems and for systems which require balanced I/O performance over multiple LUNs and I/O controllers. Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 214,21 2567,79 126,44 25832 1272 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 178,62 2579,02 0,00 25816 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 157,64 1870,13 0,00 18720 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 138,36 1677,52 0,00 16792 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 117,38 1397,00 3396,60 13984 34000 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 143,91 1688,62 0,00 16920 0 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 149,50 1868,00 252,00 18680 2520 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 126,67 1469,73 979,82 14712 9808 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 135,86 501,90 1353,05 5024 13544 Device: tps Blk_read/s Blk_wrtn/s Blk_read Blk_wrtn sda 132,87 401,20 1293,11 4016 12944 -- Tomasz Chmielewski http://wpkg.org ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ BackupPC-users mailing list [email protected] List: https://lists.sourceforge.net/lists/listinfo/backuppc-users Wiki: http://backuppc.wiki.sourceforge.net Project: http://backuppc.sourceforge.net/
