Re: ide drives performance issues, maybe related with buffer cache.

Re: ide drives performance issues, maybe related with buffer cache.

[John Newbie]

>From: "T'aZ" <taz-007@skynet.be>
>To: "John Newbie" <john_r_newbie@hotmail.com>
>Subject: Re: ide drives performance issues, maybe related with buffer 
>cache.
>Date: Sat, 16 Aug 2003 15:01:33 +0200

>RTFM , and STOP SPAMMING

I've read them, see my previous posts. If you have nothing to say, better 
stay silent.
Ok, have a look at 2.6-test2, the same picture, slightly faster, though.
And 2.6 by default set unmaskirq to 1.
IO is really bursty (see post by insecure) opposite to `enemy I`.

_________________________________________________________________
Help STOP SPAM with the new MSN 8 and get 2 months FREE*  
http://join.msn.com/?page=features/junkmail

Re: ide drives performance issues, maybe related with buffer cache.

[Alan Cox]

On Sad, 2003-08-16 at 13:35, John Newbie wrote:
> Drives were on different ide channels, but it makes no difference, even when 
> they were on the
> same channel.
> In f##king windows copying is smooth (& faster then in Linux) in both cases.

Works fine for me. But then with a name like "John Newbie" I can't help
smelling a troll 8)

Alan

Re: ide drives performance issues, maybe related with buffer cache.

[John Newbie]

Drives were on different ide channels, but it makes no difference, even when 
they were on the
same channel.
In f##king windows copying is smooth (& faster then in Linux) in both cases.

Ok, guys you may close your eyes to such a problems with end users,
but in this way you never win desktop positions.

_________________________________________________________________
STOP MORE SPAM with the new MSN 8 and get 2 months FREE* 
http://join.msn.com/?page=features/junkmail

Re: ide drives performance issues, maybe related with buffer cache.

[insecure]

On Friday 15 August 2003 18:39, John Newbie wrote:
> >How do you copy files? cp? dd? Midnight Commander? ;)

Use dd if you want to eliminate 'small buffer' problem.
With dd you can control buffer size with bs=NNN.
I found that mc is rather suboptimal...

OTOH kernel have to take care of that...
Hmm let me test it on my 2.4.21 box:

hda: 234441648 sectors (120034 MB) w/2048KiB Cache, CHS=14593/255/63, UDMA(66)
hdc: 87930864 sectors (45021 MB) w/2048KiB Cache, CHS=87233/16/63, UDMA(33)

# hdparm /dev/hda;hdparm /dev/hdc
/dev/hda:
 multcount    = 16 (on)
 I/O support  =  1 (32-bit)
 unmaskirq    =  1 (on)
 using_dma    =  1 (on)
 keepsettings =  0 (off)
 nowerr       =  0 (off)
 readonly     =  0 (off)
 readahead    =  0 (off)
 geometry     = 14593/255/63, sectors = 234441648, start = 0

/dev/hdc:
 multcount    = 16 (on)
 I/O support  =  1 (32-bit)
 unmaskirq    =  1 (on)
 using_dma    =  1 (on)
 keepsettings =  0 (off)
 nowerr       =  0 (off)
 readonly     =  0 (off)
 readahead    =  8 (on)
 geometry     = 5473/255/63, sectors = 87930864, start = 0

# hdparm -Tt /dev/hda
/dev/hda:
 Timing buffer-cache reads:   128 MB in  0.88 seconds =145.45 MB/sec
 Timing buffered disk reads:  64 MB in  2.33 seconds =27.47 MB/sec

# hdparm -Tt /dev/hdc
/dev/hdc:
 Timing buffer-cache reads:   128 MB in  0.92 seconds =139.13 MB/sec
 Timing buffered disk reads:  64 MB in  2.91 seconds =21.99 MB/sec

Copying ~600M iso from hdc to hda, Midnight Commander:

# nmeter t4 b
21:17:41.0126 bio    0    0
21:17:42.0125 bio    0    0
21:17:43.0125 bio    0    0
21:17:44.0124 bio  32k    0
21:17:45.0124 bio 7.8M 204k
21:17:46.0127 bio 9.5M    0
21:17:47.0122 bio  12M    0
21:17:48.0122 bio  12M    0
21:17:49.0121 bio 7.0M    0
21:17:50.0121 bio 7.5M 216k
21:17:51.0120 bio  15M 4096
21:17:52.0120 bio  13M  13M
21:17:53.0119 bio  10M  15M
21:17:54.0118 bio 9.5M  16M
21:17:55.0118 bio 9.7M  15M
21:17:56.0118 bio 9.5M 7.1M
21:17:57.0116 bio 9.5M    0
21:17:58.0117 bio  12M  10M
21:17:59.0115 bio 9.8M  15M
21:18:00.0179 bio  12M  13M
21:18:01.0114 bio 5.1M  18M
21:18:02.0114 bio  10M 148k
21:18:03.0113 bio  10M  72k
21:18:04.0118 bio  14M 9.9M
21:18:05.0112 bio  11M  16M
21:18:06.0112 bio 8.3M  17M
21:18:07.0111 bio 5.1M  12M
21:18:08.0112 bio  14M    0
21:18:09.0110 bio  11M 9.9M
21:18:10.0111 bio  11M  13M
21:18:11.0109 bio 9.5M  17M
21:18:12.0108 bio 6.6M  16M
21:18:13.0110 bio 8.2M    0
21:18:14.0107 bio  10M    0
21:18:15.0107 bio 9.3M 9.9M
21:18:16.0108 bio 3.2M  22M
21:18:17.0105 bio  12M 736k
21:18:18.0106 bio  13M    0
21:18:19.0104 bio  10M  17M
21:18:20.0100 bio 8.6M  26M
21:18:21.4004 bio 8.7M 4096
21:18:22.0104 bio 6.1M 212k
21:18:23.0098 bio 8.8M    0
21:18:24.0198 bio  13M  13M
21:18:25.0201 bio 9.0M  20M
21:18:26.0097 bio  13M  13M
21:18:27.0196 bio  15M  16M
21:18:28.0195 bio 7.0M  12M
21:18:29.0195 bio  16M 640k
21:18:30.0200 bio  16M  11M
21:18:31.0094 bio  15M  16M
21:18:32.0193 bio 8.3M  20M
21:18:33.0192 bio 4.5M  14M
21:18:34.0192 bio  17M    0
21:18:35.0191 bio  13M  14M
21:18:36.0191 bio  10M  17M
21:18:37.0190 bio 5.6M  14M
21:18:38.0190 bio  10M 236k
21:18:39.0189 bio 3.0M    0
21:18:40.0190 bio 3.6M    0
21:18:41.0188 bio 3.3M    0
21:18:42.0188 bio 3.7M  19M
21:18:43.0187 bio 3.5M  12M
21:18:44.0186 bio 3.0M    0
21:18:45.0186 bio 3.0M    0
21:18:46.0186 bio  10M 1.8M
21:18:47.0185 bio  15M 9.9M
21:18:48.0184 bio  11M  17M
21:18:49.0184 bio 3.7M  16M
21:18:50.0183 bio  13M 400k
21:18:51.0183 bio  13M  10M
21:18:52.0182 bio 6.9M  19M
21:18:53.0181 bio 180k 8.2M
21:18:54.0181 bio 4096 236k
21:18:55.0180 bio    0    0
21:18:56.0179 bio    0    0
21:18:57.0179 bio    0    0
21:18:58.0179 bio    0    0

Ok, now with dd bs=50M:

# nmeter t4 b
21:21:26.0096 bio    0    0
21:21:27.0200 bio    0    0
21:21:28.0095 bio 1.7M    0
21:21:29.0195 bio 8.6M    0
21:21:30.0194 bio 9.0M    0
21:21:31.0194 bio  14M 352k
21:21:32.0276 bio  11M    0
21:21:33.0093 bio 4.7M 4096
21:21:34.0204 bio 2.5M 760k
21:21:35.0091 bio 9.7M    0
21:21:36.0191 bio  16M 248k
21:21:37.0190 bio  15M 1.8M
21:21:38.0198 bio 5.7M 7.1M
21:21:39.0189 bio 5.3M  20M
21:21:40.0233 bio  10M  14M
21:21:41.0094 bio  10M  10M
21:21:42.0243 bio  10M  16k
21:21:43.0091 bio  11M  20k
21:21:44.0222 bio 2.2M  12M
21:21:45.0090 bio 7.3M  18M
21:21:46.0205 bio  12M  16M
21:21:47.0088 bio  10M 5.7M
21:21:48.0188 bio  11M  28k
21:21:49.0187 bio 9.2M 1.4M
21:21:50.0191 bio 2.3M  18M
21:21:51.0186 bio  11M  15M
21:21:52.0187 bio  11M  14M
21:21:53.0185 bio  12M 2.0M
21:21:54.0196 bio  12M  48k
21:21:55.0184 bio 385k  18M
21:21:56.0185 bio  12M  16M
21:21:57.0183 bio  10M  15M
21:21:58.0184 bio  11M 420k
21:21:59.0182 bio  10M    0
21:22:00.0222 bio 5.0M 1.3M
21:22:01.0081 bio 136k  22M
21:22:02.0181 bio 2.2M  17M
21:22:03.0180 bio  10M 316k
21:22:04.0181 bio  11M 4096
21:22:05.0179 bio  12M    0
21:22:06.0179 bio  12M 8.8M
21:22:07.0178 bio 1.7M  12M
21:22:08.0177 bio    0  25M
21:22:09.0176 bio 5.2M 8.2M
21:22:10.0176 bio  10M    0
21:22:11.0176 bio 9.1M    0
21:22:12.0176 bio  10M 108k
21:22:13.0174 bio 9.2M 4.0M
21:22:14.0193 bio 6.2M 4.2M
21:22:15.0175 bio 8.7M  20M
21:22:16.0174 bio  13M  16M
21:22:17.0172 bio  16M  10M
21:22:18.0183 bio  11M 108k
21:22:19.0171 bio    0  24M
21:22:20.0175 bio 128k  17M
21:22:21.0170 bio  16M    0
21:22:22.0170 bio  17M    0
21:22:23.0170 bio  16M 7.0M
21:22:24.0168 bio 100k  25M
21:22:25.0168 bio  24k  24M
21:22:26.0175 bio  32k  10M
21:22:27.0167 bio  12M  12k
21:22:28.0257 bio  16M  10M
21:22:29.0066 bio  14M  14M
21:22:30.0165 bio 5.5M  19M
21:22:31.0164 bio 296k 364k
21:22:32.0164 bio  14M    0
21:22:33.0163 bio 2.6M    0
21:22:34.0163 bio 3.2M    0
21:22:35.0163 bio 4.3M    0
21:22:36.0163 bio 3.3M 172k
21:22:37.0164 bio 3.5M    0
21:22:38.0164 bio 3.2M 4096
21:22:39.0160 bio 2.8M    0
21:22:40.0160 bio 8.8M    0
21:22:41.0159 bio 3.6M 208k
21:22:42.0159 bio 7.6M  22M
21:22:43.0158 bio  15M  14M
21:22:44.0167 bio  13M  11M
21:22:45.0157 bio  13M 180k
21:22:46.0158 bio  21k  13M
21:22:47.0156 bio 5.1M  18M
21:22:48.0156 bio 7.5M  20M
21:22:49.0155 bio 100k  11M
21:22:50.0154 bio    0    0
21:22:51.0154 bio 4096 204k
21:22:52.0153 bio    0    0
21:22:53.0153 bio    0    0
21:22:54.0152 bio  16k    0
21:22:55.0152 bio    0    0
21:22:56.0151 bio  32k 116k

Hmm... lets try dd if=/dev/hdc of=dummy.iso count=12 bs=50M
in order to rule out 'source file is fragmented' theory...

# nmeter t4 b
21:24:45.0090 bio    0    0
21:24:46.0198 bio    0    0
21:24:47.0193 bio    0    0
21:24:48.0193 bio    0 244k
21:24:49.0195 bio    0    0
21:24:50.0192 bio    0    0
21:24:51.0191 bio 9.6M    0
21:24:52.0191 bio  21M    0
21:24:53.4651 bio  18M 884k
21:24:54.0486 bio 6.6M 520k
21:24:55.0185 bio  22M    0
21:24:56.0211 bio  20M    0
21:24:57.0091 bio  10k  18M
21:24:58.0183 bio  10M  14M
21:24:59.0183 bio  11M  11M
21:25:00.0182 bio  11M 7.9M
21:25:01.0182 bio  16M 0.9M
21:25:02.0190 bio  16k  14M
21:25:03.0181 bio 9.1M  16M
21:25:04.0181 bio 5.5M  22M
21:25:05.2880 bio  22M 292k
21:25:06.0223 bio  13M    0
21:25:07.0084 bio 1024  20M
21:25:08.0189 bio 3.5M  13M
21:25:09.0182 bio  21M    0
21:25:10.0182 bio  23M 4.2M
21:25:11.0198 bio 1.5M  15M
21:25:12.0180 bio 2.7M  20M
21:25:13.0180 bio  10M  11M
21:25:14.0212 bio  11M  12M
21:25:15.0083 bio  14M 6.0M
21:25:16.0204 bio  10M 6.6M
21:25:17.0080 bio 5.7M  18M
21:25:18.0173 bio  11M 9.9M
21:25:19.0172 bio 4.1M  17M
21:25:20.0172 bio  21M 336k
21:25:21.0204 bio 7.1M 5.9M
21:25:22.0077 bio    0  27M
21:25:23.0255 bio  11M  13M
21:25:24.0072 bio  16M 3.8M
21:25:25.0169 bio  21M 220k
21:25:26.0187 bio 642k  16M
21:25:27.0168 bio  10M  15M
21:25:28.0168 bio  11M  11M
21:25:29.0168 bio  11M 7.6M
21:25:30.0184 bio  16M 252k
21:25:31.0167 bio 2.6M  20M
21:25:32.0167 bio  11M  11M
21:25:33.0165 bio  10M  13M
21:25:34.0168 bio  17M 4.1M
21:25:35.0164 bio 7.6M 7.7M
21:25:36.0265 bio    0  24M
21:25:37.0163 bio  13M    0
21:25:38.0163 bio  22M 116k
21:25:39.0162 bio  14M  10M
21:25:40.0162 bio 4096  25M
21:25:41.0161 bio 1024  26M
21:25:42.0161 bio  14M 6.4M
21:25:43.0160 bio  20M  52k
21:25:44.0177 bio  15M    0
21:25:45.0160 bio 1024  26M
21:25:46.0158 bio 113k 7.1M
21:25:47.0158 bio 4096 236k
21:25:48.0158 bio    0    0
21:25:49.0159 bio 180k    0
21:25:50.0157 bio    0    0
21:25:51.0156 bio    0    0
21:25:52.0156 bio    0  68k
21:25:53.0155 bio 236k    0
21:25:54.0154 bio  68k    0

What can I say? You're rught, it's bursty and not quite
up to max speed...
-- 
vda

Re: ide drives performance issues, maybe related with buffer cache.

[John Newbie]

> > So question is : why when i am copying file from one HD to another (for
> > simplicity from /hda to /hdb)
> > the speed fall down ? Starting from about 27-30 MB/s (drives are in 
>UDMA-4,
> > hdparm -X68) it drops
> > down to 11-12 MB/s after 4-5s. In *indows transfer rate is almost 
>constant
> > and about 20-22 MB/s (same hardware). Why the h#ll we suck?
> > I feel that it's due to buffer cache, because when you use sync (while
> > copying) transfer rate is so small or even 0.
> > Drives are tuned with hdparm to highest transfer rates, readahead, 
>multiple
> > sector count (hdparm
> > for details).
> > Tried different filesystems, from classic ext2/3 to modern xfs/reiserfs. 
>The
> > same results.
> > Pure kernel from kernel.org (2.4.{19,20,21}), vendors kernels - all the

>How do you copy files? cp? dd? Midnight Commander? ;)
>Does it happen with SCSI?
>--
>vda

I've used cp & Midnight Commander (mc). Also when someone uploads big file 
on server through
samba, speed sometimes fall down to zero.

Have no idea about scsi, drives are IDE.

_________________________________________________________________
Help STOP SPAM with the new MSN 8 and get 2 months FREE*  
http://join.msn.com/?page=features/junkmail

ide drives performance issues, maybe related with buffer cache.

[John Newbie]

Hi guys!

Sorry for my post, but i didnt recieve answer on it after some googling & 
posting to local Linux Users Group, and i am confident that it is related 
with kernel.
I am using linux quite long, tried (&tired of) many FS's on different 
hardware, and think this behavior is common.
So question is : why when i am copying file from one HD to another (for 
simplicity from /hda to /hdb)
the speed fall down ? Starting from about 27-30 MB/s (drives are in UDMA-4, 
hdparm -X68) it drops
down to 11-12 MB/s after 4-5s. In *indows transfer rate is almost constant 
and about 20-22 MB/s (same hardware). Why the h#ll we suck?
I feel that it's due to buffer cache, because when you use sync (while 
copying) transfer rate is so small or even 0.
Drives are tuned with hdparm to highest transfer rates, readahead, multiple 
sector count (hdparm
for details).
Tried different filesystems, from classic ext2/3 to modern xfs/reiserfs. The 
same results.
Pure kernel from kernel.org (2.4.{19,20,21}), vendors kernels - all the 
same.
Doing experiments with 'sysctl -a |grep vm' values didnt resolve the 
problem.
This behavior is general, i think.
And, damn, this is very annoying for end users, for example when copying 
large files (movies,iso's).
Beyond any doubt this is very bad for IDE based file servers.

So please help me, and sorry if this post is offtopic.

==
Linux needs grouppies !
(inspired by Almost Famous & recent akpm interview)

_________________________________________________________________
Help STOP SPAM with the new MSN 8 and get 2 months FREE*  
http://join.msn.com/?page=features/junkmail