Hi everyone, I have a board based on the SIS550 (Vortex86 @167 MHz) and I wantd to setup a NFS file server that is capable to sustain 10 MBytes/s. But the problem is that the IDE interface seems very slow in writing. I patch my kernel (2.4.19) with the patch for the 5513 and now the controller is recognise as a ATA-100a but when I try to put the controller un UDMA mode mode 5, I got a message from hdparm that told that it is not supported.
Anyway, with the UDMA mode 2 I should have the performance that I need. I look on header of the ide connector and the burst when writing seems to be quick anough but between each of these bursts I have a hole of 5-6 seconds.
I test the speed with theses commands:
time dd if=/dev/zero of=zero bs=1k count=100k -> WRITING -> 30 sec
time dd if=zero of=/dev/null bs=1k count=100k -> READING -> 10 sec
I try all configuration possible (hdparm) for the harddisk and 3 MB/s is the best result that I got. In PIO mode 4, I only got a 1.5 MB/s ! I performed theses tests on different architecture and I got the following results: SC520 PIO (ISA) -> 1.63 MB/s SC1210 PIO -> 4.54 MB/s -> I was unable to enable the DMA ???? P133 ATA-33 -> 5.50 MB/s PII-266 ATA-33 -> 17.83 MB/s PII-400 ATA-33 -> 20.54 MB/s C3-667 ATA-100 -> 23.05 MB/s
Does anyone have an idea why: 1. I cannot enable the UDMA mode 5 ? 2. I only got 3 MBytes/s in writing ?
Thanks !
Yves
Have you tried hdparm /dev/whateverX (or hdparm -t /dev/whateverX)? This will give you a more accurate result (IMHO) and it will allow you to set the DMA settings on the drive.
-CB
Thursday, October 3, 2002, 8:19:00 AM, you wrote:
YG> Hi everyone, YG> I have a board based on the SIS550 (Vortex86 @167 MHz) and I wantd to YG> setup a NFS file server that is capable to sustain 10 MBytes/s. But the YG> problem is that the IDE interface seems very slow in writing. I patch my YG> kernel (2.4.19) with the patch for the 5513 and now the controller is YG> recognise as a ATA-100a but when I try to put the controller un UDMA mode YG> mode 5, I got a message from hdparm that told that it is not supported.
YG> Anyway, with the UDMA mode 2 I should have the performance that I need. I YG> look on header of the ide connector and the burst when writing seems to be YG> quick anough but between each of these bursts I have a hole of 5-6 seconds.
YG> I test the speed with theses commands:
YG> time dd if=/dev/zero of=zero bs=1k count=100k -> WRITING ->> 30 sec
YG> time dd if=zero of=/dev/null bs=1k count=100k -> READING ->> 10 sec
YG> I try all configuration possible (hdparm) for the harddisk and 3 MB/s is YG> the best result that I got. In PIO mode 4, I only got a 1.5 MB/s ! YG> I performed theses tests on different architecture and I got the following YG> results: YG> SC520 PIO (ISA) -> 1.63 MB/s YG> SC1210 PIO -> 4.54 MB/s -> I was unable to enable the DMA ???? YG> P133 ATA-33 -> 5.50 MB/s YG> PII-266 ATA-33 -> 17.83 MB/s YG> PII-400 ATA-33 -> 20.54 MB/s YG> C3-667 ATA-100 -> 23.05 MB/s
YG> Does anyone have an idea why: YG> 1. I cannot enable the UDMA mode 5 ? YG> 2. I only got 3 MBytes/s in writing ?
YG> Thanks !
YG> Yves
YG> _______________________________________________ YG> Linuxbios mailing list YG> Linuxbios@clustermatic.org YG> http://www.clustermatic.org/mailman/listinfo/linuxbios
At 10:50 AM 10/3/02 -0400, you wrote:
Have you tried hdparm /dev/whateverX (or hdparm -t /dev/whateverX)? This will give you a more accurate result (IMHO) and it will allow you to set the DMA settings on the drive.
Yes, I do. And the results are very close to those that I got with dd. The problem whith hdparm is that it only gives the performance in reading not in writing.
I found why I cannot set the UDMA mode 3/4/5. My board has a 44 pin ide connector and the cable that come with it convert from 44 pins to a 40 pins. To be able to set the UDMA mode 3/4/5, I must used a 80 wires cable. I try on another "slow" PC and there not a great difference of performance between the UDMA mode 2 and the UDMA mode 5.
Yves
On Thursday, October 3, 2002, at 10:05 AM, Yves Godin wrote:
At 10:50 AM 10/3/02 -0400, you wrote:
Have you tried hdparm /dev/whateverX (or hdparm -t /dev/whateverX)? This will give you a more accurate result (IMHO) and it will allow you to set the DMA settings on the drive.
Yes, I do. And the results are very close to those that I got with dd. The problem whith hdparm is that it only gives the performance in reading not in writing.
try hdparm -Tt device for both read & writes.
I found why I cannot set the UDMA mode 3/4/5. My board has a 44 pin ide connector and the cable that come with it convert from 44 pins to a 40 pins. To be able to set the UDMA mode 3/4/5, I must used a 80 wires cable. I try on another "slow" PC and there not a great difference of performance between the UDMA mode 2 and the UDMA mode 5.
Yves
Linuxbios mailing list Linuxbios@clustermatic.org http://www.clustermatic.org/mailman/listinfo/linuxbios
Yves Godin wrote:
Anyway, with the UDMA mode 2 I should have the performance that I need. I look on header of the ide connector and the burst when writing seems to be quick anough but between each of these bursts I have a hole of 5-6 seconds.
How are you looking at the ide header? Logic Analyzer, scope?? If so, are you saying that the dma burst timing looks good but you are getting pauses of 5 -6 seconds between bursts?
Bari
At 01:24 PM 10/3/02 -0500, you wrote:
Yves Godin wrote:
Anyway, with the UDMA mode 2 I should have the performance that I need. I look on header of the ide connector and the burst when writing seems to be quick anough but between each of these bursts I have a hole of 5-6 seconds.
How are you looking at the ide header? Logic Analyzer, scope?? If so, are you saying that the dma burst timing looks good but you are getting pauses of 5 -6 seconds between bursts?
I`am looking the DRQ signal on th ide header with a scope. Now I'm getting pauses of 2-3 seconds in writing. It`s weird because I don`t observe this on a P133 which is slower that the SIS550 and the read seems ok.
Do you have a board with a SIS550 to compare my results ?
Yves
On Fri, 2002-10-04 at 05:36, Yves Godin wrote:
At 01:24 PM 10/3/02 -0500, you wrote:
Yves Godin wrote:
Anyway, with the UDMA mode 2 I should have the performance that I need. I look on header of the ide connector and the burst when writing seems to be quick anough but between each of these bursts I have a hole of 5-6 seconds.
How are you looking at the ide header? Logic Analyzer, scope?? If so, are you saying that the dma burst timing looks good but you are getting pauses of 5 -6 seconds between bursts?
I`am looking the DRQ signal on th ide header with a scope. Now I'm getting pauses of 2-3 seconds in writing. It`s weird because I don`t observe this on a P133 which is slower that the SIS550 and the read seems ok.
What boad and chipset revision are you using ?? Can you contact our marketing/technical support ??
Ollie
At 11:03 AM 10/7/02 +0800, you wrote:
On Fri, 2002-10-04 at 05:36, Yves Godin wrote:
At 01:24 PM 10/3/02 -0500, you wrote:
Yves Godin wrote:
Anyway, with the UDMA mode 2 I should have the performance that I
need. I
look on header of the ide connector and the burst when writing seems to be quick anough but between each of these bursts I have a hole of 5-6
seconds.
How are you looking at the ide header? Logic Analyzer, scope?? If so, are you saying that the dma burst timing looks good but you are getting
pauses
of 5 -6 seconds between bursts?
I`am looking the DRQ signal on th ide header with a scope. Now I'm getting pauses of 2-3 seconds in writing. It`s weird because I don`t observe this on a P133 which is slower that the SIS550 and the read seems ok.
What boad and chipset revision are you using ??
To evaluate the SIS550 , I'am using a board from ICOP, the ICOP-6070.
... ... SIS5513: IDE controller on PCI bus 00 dev 01 SIS5513: chipset revision 208 ...
The SIS550 is running at 167 MHz.
By using a smaller partion of type ext2 instead of ext3, a succeed to have peaks at 10 MB/s (locally) in writing with block size of 100 KBytes. I don't know if it's the smaller partition or the ext2 partition that help me. On the bigger ext3 partition, the max I got was 7 MB/s (locally).
Can you contact our marketing/technical support ??
I certainly will, but first I need to know what performance I can expected from the SIS550. I`m in a rush now, I need to determine this week if the SIS550 will be able to sustain 10 MB/s in seconds in NFS writing.
You are probably the best person to ask, what is the performance that I can expected from the SIS550 ? The 550 is our prefered choice, but if it not performant anough, we will go with the VIA C3 with a PLE133T.
Ollie
Linuxbios mailing list Linuxbios@clustermatic.org http://www.clustermatic.org/mailman/listinfo/linuxbios
On Mon, 2002-10-07 at 20:37, Yves Godin wrote:
Can you contact our marketing/technical support ??
I certainly will, but first I need to know what performance I can expected from the SIS550. I`m in a rush now, I need to determine this week if the SIS550 will be able to sustain 10 MB/s in seconds in NFS writing.
You are probably the best person to ask, what is the performance that I can expected from the SIS550 ? The 550 is our prefered choice, but if it not performant anough, we will go with the VIA C3 with a PLE133T.
I have forwarded your mail to our IAPD technical marketing personnel Jason Tsai. I believe he will give you detailed benchmark result.
Ollie
-
Bari Ari -
Christopher Bergeron -
Nathanael Noblet -
ollie lho -
Yves Godin