Re: read performance much slower than write
[email protected] (Peter T. Breuer)
| Newsgroups | gmane.linux.enbd.general |
|---|---|
| Message-ID | <[email protected]> |
Hoffmann Christian <[email protected]> wrote: > Now I started testing the performance using bonnie++ and discovered that the > write performance is much higher than the read performance! > Here is the output of bonnie++: I simpy can't read this - it looks awfully badly formatted on my 60 column screen! > bonnie++ -d /mnt/raid5/ -s 4096 -n 0 -f -u root > Version 1.03 ------Sequential Output------ --Sequential Input- --Random- > -Per Chr- --Block-- -Rewrite- -Per Chr- > --Block-- --Seeks-- > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec > %CP > enbd-client26 4G 72113 42 18330 9 > 11445 3 755.9 2 > enbd-client26 4G,,,72113,42,18330,9,,,11445,3,755.9,2,,,,,,,,,,,,, Uh, let me try and guess what it says ... Sequential output? byte and block? > Machine Size K/sec %CP K/sec %CP > enbd-client26 4G 72113 42 18330 9 Well, that's 18MB/s block and 72MB/s char! I really am not sure I believe that! > enbd-client26 4G,,,72113,42,18330,9,,,11445,3,755.9,2,,,,,,,,,,,,, And I don't know what that is. > As you can see the write performance is very high, the read performance is > even worse than the rewrite. Well, I simply can't read it! So I don't know what it says. But if it says that writing char by char is 72MB/s and block by block is 18MB/s, then I am already puzzled. > I already tried to echo "merge_requests=512" > /proc/nbdinfo which didn?t > change much. > Shouldn?t be the performance of read operations better or at least almost > the same than the write performance? It should be about the same. On write one can save quite a bit through md5summing (both sides) if one is always rewriting the same stuff, but it should be at most a 2:1 gain (approx). Measured read perfomance is generally always higher than write, although I am not sure exactly why (without thinking about it). Are your logs full of some complaints? Peter