Quick navigation:        Home   |    Site Map   ||    References   |    Biography   ||    Copyright   |    Other copyright   |    Contact us   |   
Protein structure
 

Re: [ccp4bb] raid array load comments
 

Basic tutorials: 		 
 

CCP4bb navigation

CCP4bb <-- 2008 <-- January 2008 <-- 15 January 2008
Previous message:
Subject: Re: wich program to test an alternative phaser solution
From: Airlie McCoy ajm201 {- at -} CAM {- dot -} AC {- dot -} UK
Date: 2008-01-15 		Next message:
Subject: Re: PERL system call to CCP4
From: Jawahar Swaminathan jawahar {- at -} EBI {- dot -} AC {- dot -} UK
Date: 2008-01-15


Subject: Re: raid array load comments
From: Kay Diederichs Kay {- dot -} Diederichs {- at -} UNI-KONSTANZ {- dot -} DE
Date: 2008-01-15
Harry M. Greenblatt schrieb:
> BS"D
>
> Thank you to all the respondents.
>
> Some comments:
>
> 1. Some believe that the write performance in RAID 5 is only as good as

I'm one of those "some" (based on experience, and a basic understanding
of how RAID5 operates).

Even 3ware gives performance data for the 9550SX controller
()
which show that write performance is only about half of read performance
in RAID5 - in an optimized situation.

> performance to one disk. This is true only in RAID 3 (under certain
> conditions), where parity is written as a separate operation to one
> dedicated parity disk. With distributed parity in RAID 5 this
> bottleneck is overcome, and write performance is better than single disk
> limits.
>
> 2. These are SATA II drives which are rated at 3Gb/s, or 375MB/s; even
> if single-drive I/O limits were a problem, we are not reaching this value.
>

These numbers refer to the SATA II protocol in general, and are an upper
limit for the actual speed of transfer between spindle and operating
system (i.e. when you transfer only between disk cache and controller,
these numbers may be almost reached in favourable cases). However, if
the data are on the disk platters or should be written to them, the
speed will be on the order of 60 MB/s for these disks.

A simple way to come up with realistic upper limits is to run
"hdparm -tT /dev/sdX" where X addresses a single disk or RAID array. The
first line of output gives the interface speed, and the second the
spindle speed. For your application, the latter is the important one.

Furthermore, the combination of 3ware controllers, RAID5 and ext3 file
systems is known to be slow. XFS should be better, and that's what 3ware
recommends for Linux. If you want to try XFS: CentOS has XFS kernel
modules and tools. If you want to stay with ext3, make sure you have the
latest kernel modules and tools from the 3ware site, and the latest
firmware for the controller.

HTH,

Kay
--
Kay Diederichs http://strucbio.biologie.uni-konstanz.de
email: Kay.Diederichs@uni-konstanz.de Tel +49 7531 88 4049 Fax 3183
Fachbereich Biologie, Universität Konstanz, Box M647, D-78457 Konstanz
ProteinCrystallography.org: Copyright 2006-2007 by Quid United Ltd