Gentoo Archives: gentoo-user

From: Kellystewart00@×××××××××.au
To: gentoo-user@l.g.o
Subject: Re: [gentoo-user] Performance problem at writing big files and Multitasking
Date: Tue, 27 Feb 2007 07:56:04
Message-Id: 1623124876-1172562578-cardhu_decombobulator_blackberry.rim.net-2082807461-@bxe008-cell00.bisx.prodap.on.blackberry
In Reply to: Re: [gentoo-user] Performance problem at writing big files and Multitasking by Matthias Bethke
1 Sent via BlackBerry® from Vodafone
2
3 -----Original Message-----
4 From: Matthias Bethke <matthias@×××××××.de>
5
6 Date: Mon, 26 Feb 2007 17:28:35
7 To:gentoo-user@l.g.o
8 Subject: Re: [gentoo-user] Performance problem at writing big files and Multitasking
9
10
11 Hi Daniel,
12 on Saturday, 2007-02-10 at 12:49:14, you wrote:
13 > I will give short overview what i have tried so far.
14 >
15 > 1. Trying different I/O Scheduler ( cfq anticipatory and deadline)
16 > 2. Enabling Low latency kernel and Preemptible kernel
17 > 3. Setting 1000 HZ for timer frequency
18 > 4. Tried the new kernel 2.6.19-gentoo-r6 and even the testing version
19 > 2.6.20-gentoo with core 2 enabled in processor type
20
21 Oh, so it is a multicore CPU---sorry if you mentioned it already, I
22 had deleted the start of the thread already when I read Benno's advice.
23 In that case, try 100 Hz scheduling period as well. I've had very bad
24 experiences with I/O and 250 Hz or higher on a dual Xeon. My guess is
25 that it was a cache effect and therefore shouldn't happen on the
26 Core2Duo, but it might still be worth a try.
27
28 > As i am using Xfce i installed the diskperf-plugin which monitors disk
29 > I/O. The monitoring is divided in disk-read and disk-write.
30 > I recognized that every time when reading stops writing starts. So is
31 > this staggering of writing to disk normal as the programs have to read
32 > data they want to write to disk? On my previous machine i didn't
33 > recognize such a behaviour.
34
35 So you're reading and writing from/to the same disk? I'd expect that
36 behavior then, because the I/O scheduler tries to satisfy requests with
37 as little thrashing as possible. So if there are enough write requests
38 queued up it may keep the HD busy writing for a while before reading the
39 next chunk from somewhere else.
40
41 cheers!
42 Matthias
43 --
44 I prefer encrypted and signed messages. KeyID: FAC37665
45 Fingerprint: 8C16 3F0A A6FC DF0D 19B0 8DEF 48D9 1700 FAC3 7665
46
47 éí¢‹¬z¸žÚ(¢¸&j)bž b²