Hi everyone

OS: Mandrake 10.1 Official

I've built a Fileserver with 11 IDE Drives, 3 drives on the Integrated IDE Controller, 8 on two Promise U133TX2 PCI IDE Controllers.

All disks are Maxtor 200GB Drives (6Y200P0).

My Problem is, all Slave Drives have a very low Performance:

/dev/hda: Timing buffered disk reads: 128 MB in 3.01 seconds = 42.47 MB/sec
/dev/hdb: Timing buffered disk reads: 8 MB in 3.23 seconds = 2.47 MB/sec
/dev/hdc: Timing buffered disk reads: 134 MB in 3.03 seconds = 44.27 MB/sec
/dev/hdd: Timing buffered disk reads: 8 MB in 3.23 seconds = 2.48 MB/sec
/dev/hde: Timing buffered disk reads: 132 MB in 3.01 seconds = 43.85 MB/sec
/dev/hdf: Timing buffered disk reads: 8 MB in 3.23 seconds = 2.48 MB/sec
/dev/hdg: Timing buffered disk reads: 106 MB in 3.03 seconds = 34.93 MB/sec
/dev/hdh: Timing buffered disk reads: 8 MB in 3.23 seconds = 2.48 MB/sec
/dev/hdi: Timing buffered disk reads: 86 MB in 3.04 seconds = 28.28 MB/sec
/dev/hdk: Timing buffered disk reads: 86 MB in 3.04 seconds = 28.29 MB/sec
/dev/hdl: Timing buffered disk reads: 86 MB in 3.03 seconds = 28.39 MB/sec

hda - h are drives on a PCI controller, the other ones are on the onboard IDE. All drives are connected with the same cables. All drives are CS Jumpered (more speed than Master/Slave, don't know why )

I've tried to set the udma Mode manually to udma6 (All Drives, the Controllers and the Cables (80pin) should support that). But a few moments later, the os sets the master drives to udma4, dma enabled and the slave drives to udma6, dma disabled, running in pio mode 0. setting all drives to udma4 won't work, either.

what should i do?

Thanks a lot to everyone that attempts to help!

PS: I have another fileserver, same disks, same controllers but mandrake 9.0. output of hdparm:

/dev/hda: Timing buffered disk reads: 64 MB in 2.71 seconds = 23.62 MB/sec
/dev/hdb: Timing buffered disk reads: 64 MB in 2.04 seconds = 31.37 MB/sec
/dev/hdc: Timing buffered disk reads: 64 MB in 2.09 seconds = 30.62 MB/sec
/dev/hdd: Timing buffered disk reads: 64 MB in 2.68 seconds = 23.88 MB/sec
/dev/hde: Timing buffered disk reads: 64 MB in 1.34 seconds = 47.76 MB/sec
/dev/hdf: Timing buffered disk reads: 64 MB in 1.32 seconds = 48.48 MB/sec
/dev/hdg: Timing buffered disk reads: 64 MB in 1.31 seconds = 48.85 MB/sec
/dev/hdh: Timing buffered disk reads: 64 MB in 1.31 seconds = 48.85 MB/sec
/dev/hdi: Timing buffered disk reads: 64 MB in 1.26 seconds = 50.79 MB/sec
/dev/hdj: Timing buffered disk reads: 64 MB in 1.31 seconds = 48.85 MB/sec
/dev/hdk: Timing buffered disk reads: 64 MB in 1.46 seconds = 43.84 MB/sec
/dev/hdl: Timing buffered disk reads: 64 MB in 1.32 seconds = 48.48 MB/sec

"I've tried to set the udma Mode manually to udma6 (All Drives, the Controllers and the Cables (80pin) should support that). But a few moments later, the os sets the master drives to udma4, dma enabled and the slave drives to udma6, dma disabled, running in pio mode 0. setting all drives to udma4 won't work, either."

Promise has a history of building buggy IDE controller chips. The kernel drivers for Promise IDE chipsets have work arounds for the Promise bugs which usually result in slow performance because the chipset cannot work error free at maximum performance. What you describe sounds like Linux running a buggy Promise chipset at the maximum error free performance level possible.

I threw away my Promise IDE controller (which was not your model) and bought High Point which works perfectly.

-------------------------
Steve Stites

I was about to put a long post up somewhere and found this first, so this is the perfect place. I have several contollers laying around, and i was wanting to get all my drives on seperate controllers for performance reasons. I've had a Highpoint 370 in my main machine for a long time, and generaly just use it to hook up an extra hard drive. I wanted to be able to hook up some extra dvd drives, but the high point doesn't really support them, and they run at PIO mode 0.

I have 2 Promise IDE controllers also. One is based on the 20267 chip and one on the 20268. I didn't realize at first that was where the kernel made the split in the drivers. 20267 and before was one driver and 20268 and after is another. I rebuilt a kernel, 2.4.23 with support for the promise 20268, and hooked up dvd drives there. They worked, and i was happy. I Tested the speed of dvd rips from each drive on the promise and on the onboard VIA 266 chipset controller, and got the same speed.

I then noticed that the hard drive hooked up to my Highpoint seemed slower. I ran hdparm and got about 28 megs a second. The drive was getting 47 before i put in the promise card. This wasn't going to cut it. I put the hard drive on the promise controller, and got 40 megs a second, but i got weird slow choppy performance copying files to that drive.
I have a setting in my rc.local.
setpci -v -H1 -s 0:0.0 70=86
setpci -v -H1 -s 0:0.0 95=1e
that enables the halt when idle, that allows my machine to run much cooler. i had noticed that this makes the highpoint run a little slower, maybe 15%, but doesn't affect the onboard controller. It was making the hard drive on the promise run very slow. Perhaps 90% slower or more than it was supposed to. hdparm would give me normal speeds, but copying files to the drive would be very slow, similar to PIO speeds. When i disabled those settings, the hard drive acted normally.

I upgraded the firmware on the promise 20268 to see if that would help. It just hurt. The dvd drives would only run in PIO mode, and the hard drive wasn't helped. The promise controller was also still interfering with the Highpoint's performance. I reflashed the old firmware.

Another weird happening, was that the promise was in a higher pci slot, but would get read first when no hard drive was attached to the highpoint, but the highpoint would get read first, if a hard drive was attached. I mean at boot where the controller's bios is reading what's hooked up to it, and in linux, which drives get what letters. The highpoint seemed to swap places with the promise, if a hard drive was hooked to the highpoint.

I then was testing the performace of the dvd drives hooked up the promise. When a single drive was ripping a dvd, performance was normal, but when 2 drives were ripping on the same cable, performance took a huge dive. Much poorer than 2 on the same cable would get on the via on board controller. Say each dvd drive would rip a dvd in 15 minutes on it's own, and they would take 25 minutes total when running together on the onboard via, but would take an hour and 15 minutes together running on the promise. I was going to have to have them on seperate cables.

I added the other promise controller, the 20267 chip based one, and removed the highpoint. I didn't realize at this point that this would take a different kernel driver. I booted with a hard drive hooked to the promise 20267, and 2 dvd drives hooked to the 20268. I ran hdparm on the letter i thought the hard drive would be at. A dvd drive light lit up, but the hard drive started running too. the drive letter was a dvd, but it was also accessing the hard drive too !!!

I got 28 megs a second, but i couldn't get to the hard drive's partition table or anything. The controller's were messing with each other. Super weird. I feared for my hard drive and shut down, and unhooked it. i then recompiled the kernel with the right drivers and things were better. I was getting good speed from the dvd drives on different cables. The hard drive on the 20267 was getting 47 megs a second again, even with my setpci commands for halt on idle, and i think the computer is stable. It never did freeze in all this weirdness.

I wouldn't have even had to mess with those crazy promise controllers if the highpoint 370 supported cdrom and dvd drives ok. I'm still not sure that my dvd drives' performance is better like this than they were on the same cable on the internal ide, but i can now run a dvd rip, while burning a dvd on a different drive on a different cable, without them messing each other up.


I also had trouble making the new devices necessary to access drive pas letter hdg. My MAKEDEV couldn't handle hdi, and i kept downloading different MAKEDEV scripts, and nothing could do it. Finally i happend upon a google post that said the script was in my kernel source. /usr/src/linux/scripts/MAKEDEV.ide was it. Worked fine.