I know this one! Got to go from memory since I can't find Neil's answer. md will do 4K disk I/O. I am not sure about the size, but much smaller than normal stripe size. If you were to change 1 sector, md would read the 4K that contains the sector, and read the 4K of parity. Factor out the old data, factor in the new data. Then write the new 4K data and 4K parity. - Get old data: 1 read (4K) - Get current/old parity: 1 read (4K) - factor out old data (XOR) - compute new parity (XOR) - Write data: 1 write (4k) - Write parity: 1 write (4k) There is some read-ahead, but I don't know what layer does that. My array is over 3 times as fast as a single disk. Using dd I get 19M/sec on any 1 disk, but 61M/sec on the array. My write speed is 30M/sec creating a file on a filesystem. Multi-threaded random read speed is much faster than 1 disk. Maybe 14 times faster, but not sure. Guy -----Original Message----- From: linux-raid-owner@xxxxxxxxxxxxxxx [mailto:linux-raid-owner@xxxxxxxxxxxxxxx] On Behalf Of Sebastien Koechlin Sent: Tuesday, September 14, 2004 10:46 AM To: Lukas Kubin Cc: Guy; linux-raid@xxxxxxxxxxxxxxx Subject: Re: Array of disks attached to multiple controllers On Tue, Sep 14, 2004 at 04:07:43PM +0200, Lukas Kubin wrote: > Chunk Size : 128K (...) > Number Major Minor RaidDevice State > 0 8 16 0 active sync /dev/sdb > 1 8 32 1 active sync /dev/sdc > 2 8 48 2 active sync /dev/sdd > 3 8 64 3 active sync /dev/sde > 4 8 80 4 active sync /dev/sdf > 5 8 96 5 active sync /dev/sdg > 6 8 112 6 active sync /dev/sdh > 7 8 128 7 active sync /dev/sdi > 8 8 144 8 active sync /dev/sdj > 9 8 160 9 active sync /dev/sdk > 10 8 176 10 active sync /dev/sdl > 11 8 192 11 active sync /dev/sdm > 12 8 208 12 active sync /dev/sdn > 13 8 224 13 active sync /dev/sdo > 14 8 240 14 active sync /dev/sdp > 15 65 0 15 active sync /dev/sdq > 16 65 16 16 spare /dev/sdr I have a question about performance: What is the cost of writing a 'data-unit' in such an array? - Write data: 1 write - Calculate new checksum: 14 reads - Write checksum: 1 write Right or wrong? What is the granularity of thoses ckecksum updates? 512 bytes (sector size)? 4k (page size on i386)? chunk-size? Does Linux do read-ahead on thoses 14 disks reads? Thanks -- Seb, autocuiseur - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
