Sequential ZFS Mirrors

Last month, an ambassador colleague shared the idea of having ZFS mirrors on only one hard drive. One of the immediate questions is: how fast is such a zpool? I did some tests on a 1.7 GHz Pentium M laptop with a 120 GB IDE drive (WDC WD1200BEVE) and 1 GB of main memory, running the OpenSolaris 2008.05 distribution. Leaving some space behind the root file system during the installation, I used the remaining three primary partitions for a zpool (“m3”) where the data was replicated twice (two mirrors).

Results

Testhomedir (within rpool)m3
Creating a 1 GB file (with dd)33 s, 32 MB/s134 s, 7.8 MB/s
Reading the 1 GB file (with dd)22 s, 46.7 MB/s41 s, 25.6 MB/s
Compiling the wxWidgets library9 m 18 s9 m 41 s

Kommentare:

This also doesn't achieve much in the way of protection. A simpler and better performing way is to set "copies=2" on the datasets that you need to "mirror". This provides basically the same level of redundancy but won't have the performance hit.

Gesendet von Darren Moffat am Juli 22, 2008 at 06:15 AM MESZ #

Does someone know how copies=2 distributes the data on the drive. We definitly want to have a seek between copies, to minimize locality of the data. This is important, because disk errors have a very high locality.

Gesendet von Marc am Juli 25, 2008 at 11:48 AM MESZ #

Today I did the tests with "copies=3" (copies=2 means single replication):

Creating a 1 GB file : 10.7 MB/s
Reading the 1 GB file : 40.7 MB/s
Compiling wxWidgets : 9 m 24 s

@Marc: http://blogs.sun.com/bill/entry/ditto_blocks_the_amazing_tape says "Our policy aims to place the blocks at least 1/8 of the disk apart."

Gesendet von Igor Geier am Juli 31, 2008 at 08:17 PM MESZ #

Senden Sie einen Kommentar:
  • HTML Syntax: Ausgeschaltet
About

Sun Campus Ambassador,
University of Frankfurt (Germany)

Search

Categories
Archives
« April 2014
MoDiMiDoFrSaSo
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
    
       
Heute