linux-kernel - Re: How do I self diagnose slow write to HDD performance issue

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <111655.1224644845@turing-police.cc.vt.edu>
Date:	Tue, 21 Oct 2008 23:07:25 -0400
From:	Valdis.Kletnieks@...edu
To:	Jonathan Johnson <johnsonn@...c.edu>
Cc:	linux-kernel@...r.kernel.org
Subject: Re: How do I self diagnose slow write to HDD performance issue

On Tue, 21 Oct 2008 14:34:01 CDT, Jonathan Johnson said:

> Before I setup the array I tested the SATA 2 drives connected to the mobo SATA connect and got approx 60mb/s
> write and even better read speeds per drive.

Remember that streaming read/write speed is usually only a small part of the
total disk performance story.  Sure, if you don't have to move the heads much,
you can go 60 megabytes/sec.  However, even the *hottest* drives out there
have a hard time sustaining 200 seeks per second.

In other words, every time you have to move the heads, it costs you 300Kbytes/sec
of throughput (1/200th of that 60M).

And bonnie++ probably moves the heads a lot.

To *add* insult to injury, if you're running in RAID6 mode, each time you write
to the RAID set, it has to do *several* I/Os to do all the parity computations.
So you take a 300K hit for each of those I/Os.

There's a *reason* why RAID1 is still used - you can (assuming a smart enough
controller) launch both the write and the mirroring write at the same time in
parallel, taking little or no performance penalty.

Content of type "application/pgp-signature" skipped