read vs write. Some tests.

read vs write. Some tests.

[Fyodor Ustinov]

ceph - 0.28.2 from repository

client - linux 2.6.39/64

All osd servers have 4G RAM. Journal on 1G tempfs. 2*1G ethernet 
(etherchannel)

client server have 2G RAM. 2*1G ethernet (etherchannel)

In tests used "dd if=/dev/zero of=file bs=8M count=3000" for write and 
"dd if=file of=/dev/null bs=8M count=3000" for read. For rbd instead 
"file" used "/dev/rbd0".
In "ceph fs" additionaly used "iozone -a -y 8M -i 0 -i 1 -i 2 -n25g -g25g"

Tests results:

kernel ceph fs:
dd write:  149.748 s, 168 MB/s
dd read:  451.835 s, 55.7 MB/s
iozone write: 165 MB/s
iozone rewrite:  165 MB/s
iozone randow write: 165 MB/s
iozone read: 53 MB/s
iozone reread: 55 MB/s
iozone random read: 41 MB/s

rbd:
dd wite: 212.583 s, 118 MB/s
dd read: 480.374 s, 52.4 MB/s


Frankly I do not quite understand what was happening.

WBR,
     Fyodor.