[prev in list] [next in list] [prev in thread] [next in thread]
List: aix-l
Subject: tuning sequential io on striped lv
From: Holger.VanKoll () SWISSCOM ! COM
Date: 2003-04-24 11:05:21
[Download RAW message or body]
[Attachment #2 (multipart/alternative)]
Hello,
from aix-docs:
Set maxpgahead to 16 times the number of disk drives (vmtune -R).
This causes page-ahead to be done in units of the stripe-unit size (64
KB) times the number of disk drives,
resulting in the reading of one stripe unit from each disk drive for
each read-ahead operation
Ok, that means with a stripe-size of 128kb I choose 32 times the number
of disk-drives, correct?
Now... I got several VG with striped FS, they contain different number
of disk-drives.
I am not sure what "number of disk drives" to choose.
8 VG have 5 drives, 2 have 4 drives, 1 has 10 drives.
If I set maxpgahead to 32*5, those 8 vg will perform well and the 1 with
10 will perform quite well, too.
If I set it to 32*10, the 8 vgs will still perform quite well and the 1
with 10 well, too. But I risk more unnecessary IO then with 32*5; and
especially on the 8VG this will increase disk-busy% (tm_acct).
So, 32*5 sounds good to me. What do you think?
But what about 32*4? The 2VG will perform well, but what about the 8VG
then? Will they perform at 4/5 as much as they would perform at 32*5 or
less?
I think it will be close to 4/5 but would like to hear your opinion.
Regards,
Holger
[Attachment #5 (text/html)]
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=us-ascii">
<META NAME="Generator" CONTENT="MS Exchange Server version 6.0.6249.1">
<TITLE>tuning sequential io on striped lv</TITLE>
</HEAD>
<BODY>
<!-- Converted from text/rtf format -->
<P><SPAN LANG="de"><FONT SIZE=2 FACE="Arial">Hello,</FONT></SPAN>
</P>
<P><SPAN LANG="de"><FONT SIZE=2 FACE="Arial">from aix-docs:</FONT></SPAN>
</P>
<P><SPAN LANG="de"><FONT SIZE=2 FACE="Courier New">Set maxpgahead to 16 times the \
number of disk drives (vmtune -R). </FONT></SPAN>
<BR><SPAN LANG="de"><FONT SIZE=2 FACE="Courier New">This causes page-ahead to be done \
in units of the stripe-unit size (64 KB) times the number of disk drives, \
</FONT></SPAN>
<BR><SPAN LANG="de"><FONT SIZE=2 FACE="Courier New">resulting in the reading of one \
stripe unit from each disk drive for each read-ahead operation</FONT></SPAN> </P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">Ok, that means with a stripe-size of \
128kb I choose 32 times the number of disk-drives, correct?</FONT></SPAN> </P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">Now... I got several VG with striped \
FS, they contain different number of disk-drives.</FONT></SPAN> </P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">I am not sure what "number of \
disk drives" to choose. </FONT></SPAN>
<BR><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">8 VG have 5 drives, 2 have 4 drives, \
1 has 10 drives.</FONT></SPAN> </P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">If I set maxpgahead to 32*5, those 8 \
vg will perform well and the 1 with 10 will perform quite well, too.</FONT></SPAN>
<BR><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">If I set it to 32*10, the 8 vgs will \
still perform quite well and the 1 with 10 well, too. But I risk more unnecessary IO \
then with 32*5; and especially on the 8VG this will increase disk-busy% \
(tm_acct).</FONT></SPAN></P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">So, 32*5 sounds good to me. What do \
you think?</FONT></SPAN> </P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">But what about 32*4? The 2VG will \
perform well, but what about the 8VG then? Will they perform at 4/5 as much as they \
would perform at 32*5 or less?</FONT></SPAN></P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">I think it will be close to 4/5 but \
would like to hear your opinion.</FONT></SPAN> </P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">Regards,</FONT></SPAN>
</P>
<P><SPAN LANG="en-us"><FONT SIZE=2 FACE="Arial">Holger</FONT></SPAN>
</P>
</BODY>
</HTML>
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic