Re: RAID 5 performance

From: George Dau <gedau_at_mim.com.au>
Date: 1995/11/18
Message-ID: <DI7tpp.IIr_at_mim.com.au>#1/1

100440.1227_at_compuserve.com (Ashley) wrote:

>RAID 5 is very, very bad if you are performing many writes, the reason

 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ b.s.

>for this is that the data is written first, then it has to be read
>back from the disks in order to create the checksum/parity data, then
>it has to write this data back to the disks as well. Thus in order to
>write data it actually performs two writes and one read. However,
>your read times should be reasonable. RAID 5 was a compromise to
>offer data integrity, and speed was never really the issue. As for
>your RAID 0+1 experiment, it should perform pretty well whilst
>offering the best integrity, I'll be very interested to see any data
>you get,
>
>Ashley
>
>

48 hrs worth of stats from our system. Mainly doing data loads.

                        OPERATIONS           BLOCKS        AVG
TIME(ms)
TYP NAME              READ     WRITE      READ     WRITE   READ  WRITE

vol swap1            42339     36264    338712    664984    4.9   15.9

vol swap2            41791     35735    337208    666728    4.8   16.7

vol swap3            41118     35631    329904    666432    4.9   16.0

vol swap4            41117     35594    328936    663728    4.9   15.4

vol swap5            41959     35441    335672    663928    4.7   16.0

vol swap6            42930     34935    343440    664504    4.8   15.9

vol swap7            41405     34917    331240    663672    4.8   15.7

vol swap8            43128     36252    345024    666176    4.6   16.5

vol u02            1127540    192568  17820432   5324936    8.5   41.2

vol u03             430586     44310   6982670    397640    3.4    1.9

vol u04             422324     16585   7034368   1672954    3.6   17.6

vol u05             129471    355269   1787168   3623226    9.4   37.0

vol u06              98854    181705   1353540   1504864   15.2   38.4

vol u07             956228    124167  22423350   9435826    4.9    4.3

vol u08             384896     10317   6179095    101876    2.2   14.8

vol u09              72962    436968   1539558   5056776   17.6    2.3

Each swap vol is mirrored (no striping)
u02 is raid over 7 disks + extra for log file u03 is striped over 8 disks
u04 is striped over 8 disks sharing disks with u05 u05 is raid 5 over 8 disks + extra for log file u06 is raid 5 over 8 disks + extra for log ifle u07 is striped over 7 disks + extra for log file u08 is messy bits of concat split all over the place u09 is is striped over 7 disks.

An interesting thing to look at is compare u02 with u07. Both have about 4 times more writes than reads. A lot of people have been knocking raid 5 with write-busy file systems.

Even with 4X write to read, raid 5 returns 41.2ms write times. Striping is 4.3ms, a saving of 38ms and a cost of 7 X 2Gig Drives. 41.2ms writes are quite fast enough for us so far, and we still have the option of getting more disks and going to stiped mirrors if we need the performace later.

Regards, George Dau
gedau_at_mim.com.au Received on Sat Nov 18 1995 - 00:00:00 CET