Hi,
just subscribed to this list also! Got no useful answers in l-k and l-raid
and found rudiments of it in l-raid. Here is the original :-)
The promise folks claim working linux support for their IDE product line.
While failed in setting up a very simple config with 2 * TX2 and UDMA (see
forwarded message below), and googling around a bit, I ask myself, how
stable/fast other PDC20268 based raid solutions are?
In other words, there are well known issues with multiple TX2 controllers.
Some are hardware limits, and some are pdc202xx.c driver issues, and it's
not sure, if/how they are solved, because Promise doesn't kept a linux
friendly company after releasing their kernel hacker Frank Tiernan.
OTOH, the values below looks quite promising. If only the sencond TX would
behave correctly. I would like to know, how other multi PDC20268 based
controllers behave in this respect, especially the TX4 (for sw raid).
Does a single TX4 suffer from the same UDMA limitations on the 3rd and 4th
channel like 2 TX2's? How does 2 TX4's behave in this respect?
I'm looking for a stable and fast sw or hw raid solution for my dual K7
thunder. The plan was using multiple simple fast ide contr. with sw raid.
Because sw raid takes cpu, I thing two K7 XP1700 should be sufficient ;-)
Tomorrow, I (hopefully) will get a TX4 and will check it then. I will let
you know...
Promise offers a i2o based hw raid controller SX6000. According to Alan Cox,
it should work. If somebody tried it, what performance can I expect from it?
BTW: Does somebody knows Andre Hedricks current email address. I found
several, but refrained from spamming them all.
Thanks in advance
Hans-Peter
---------- Forwarded Message ----------
Subject: 2nd promise Ultra100 TX2 runs degraded
Date: Mon, 22 Oct 2001 18:41:57 +0200
From: Hans-Peter Jansen <hpj@xxxxxxxxx>
To: linux-kernel@xxxxxxxxxxxxxxx
Cc: Andre Hedrick <andre@xxxxxxxxxxxxxxxxx>
Hi there,
while trying to set up a software raid 5 with 60gb ide drives,
i've encountered following problem:
the performance on the second promise controller is really bad.
HW: Tyan K7 Thunder, 2 * XP1700, 2GB, 2 * Promise Ultra100 TX2
4 * 60GB IBM IDE
OS: Linux 2.4.12
boot.msg:
<6>Uniform Multi-Platform E-IDE driver Revision: 6.31
<4>ide: Assuming 33MHz system bus speed for PIO modes; override with
idebus=xx <4>AMD7411: IDE controller on PCI bus 00 dev 39
<4>AMD7411: chipset revision 1
<4>AMD7411: not 100%% native mode: will probe irqs later
<4>AMD7411: disabling single-word DMA support (revision < C4)
<4> ide0: BM-DMA at 0xf000-0xf007, BIOS settings: hda:pio, hdb:pio
<4> ide1: BM-DMA at 0xf008-0xf00f, BIOS settings: hdc:pio, hdd:pio
<4>PDC20268: IDE controller on PCI bus 00 dev 40
<4>PDC20268: chipset revision 2
<4>PDC20268: not 100%% native mode: will probe irqs later
<4>PDC20268: (U)DMA Burst Bit DISABLED Primary PCI Mode Secondary PCI Mode.
<4>PDC20268: FORCING BURST BIT 0x00 -> 0x01 ACTIVE
<4> ide2: BM-DMA at 0x2010-0x2017, BIOS settings: hde:pio, hdf:pio
<4> ide3: BM-DMA at 0x2018-0x201f, BIOS settings: hdg:pio, hdh:pio
<4>PDC20268: IDE controller on PCI bus 00 dev 48
<4>PDC20268: chipset revision 2
<4>PDC20268: not 100%% native mode: will probe irqs later
<4>PDC20268: (U)DMA Burst Bit DISABLED Primary PCI Mode Secondary MASTER
Mode. <4>PDC20268: FORCING BURST BIT 0x50 -> 0x51 INACTIVE
<4> ide4: BM-DMA at 0x2020-0x2027 -- ERROR, PORT ADDRESSES ALREADY IN USE
<4> ide5: BM-DMA at 0x2028-0x202f -- ERROR, PORT ADDRESSES ALREADY IN USE
<4>hde: IC35L060AVER07-0, ATA DISK drive
<4>hdg: IC35L060AVER07-0, ATA DISK drive
<4>hdi: IC35L060AVER07-0, ATA DISK drive
<4>hdk: IC35L060AVER07-0, ATA DISK drive
<4>ide2 at 0x2048-0x204f,0x2042 on irq 3
<4>ide3 at 0x2038-0x203f,0x2036 on irq 3
<4>ide4 at 0x2060-0x2067,0x205a on irq 10
<4>ide5 at 0x2050-0x2057,0x2046 on irq 10
<6>hde: 120103200 sectors (61493 MB) w/1916KiB Cache, CHS=119150/16/63,
UDMA(100)
<6>hdg: 120103200 sectors (61493 MB) w/1916KiB Cache, CHS=119150/16/63,
UDMA(100)
<6>hdi: 120103200 sectors (61493 MB) w/1916KiB Cache, CHS=119150/16/63
<6>hdk: 120103200 sectors (61493 MB) w/1916KiB Cache, CHS=119150/16/63
Note: hdi and hdk run in in pio mode (forcing them to something better
with hdparm failed). During init of the 2nd card, something went wrong.
What does the burst bit mean, and why get's it disabled.
There seem to be some problems with allocation of the ioports
of the secondary card. Therefore the relevant parts of /proc/ioports:
2010-201f : Promise Technology, Inc. 20268
2010-2017 : ide2
2018-201f : ide3
2020-202f : Promise Technology, Inc. 20268 (#2)
2020-202f : PDC20268
2030-2033 : PCI device 1022:700c (Advanced Micro Devices [AMD])
2034-2037 : Promise Technology, Inc. 20268
2036-2036 : ide3
2038-203f : Promise Technology, Inc. 20268
2038-203f : ide3
2040-2043 : Promise Technology, Inc. 20268
2042-2042 : ide2
2044-2047 : Promise Technology, Inc. 20268 (#2)
2046-2046 : ide5
2048-204f : Promise Technology, Inc. 20268
2048-204f : ide2
2050-2057 : Promise Technology, Inc. 20268 (#2)
2050-2057 : ide5
2058-205b : Promise Technology, Inc. 20268 (#2)
205a-205a : ide4
2060-2067 : Promise Technology, Inc. 20268 (#2)
2060-2067 : ide4
f000-f00f : Advanced Micro Devices [AMD] AMD-765 [Viper] IDE
f000-f007 : ide0
f008-f00f : ide1
Although the driver claims problems with 0x2020-0x202f,
they're assigned to the promise driver.
For the sake of completeness: cat /proc/interrupts
CPU0 CPU1
0: 69399 96814 IO-APIC-edge timer
1: 1527 1358 IO-APIC-edge keyboard
2: 0 0 XT-PIC cascade
3: 9686 10007 IO-APIC-level ide2, ide3
5: 0 0 IO-APIC-level eth0
8: 0 0 IO-APIC-edge rtc
10: 35030 35048 IO-APIC-level ide4, ide5
11: 0 0 IO-APIC-level usb-ohci
12: 6658 3380 IO-APIC-edge PS/2 Mouse
NMI: 0 0
LOC: 166126 166124
ERR: 0
MIS: 0
The result of this:
tyrex:~ # for i in hde hdg hdi hdk; do hdparm -tT /dev/$i; done
/dev/hde:
Timing buffer-cache reads: 128 MB in 0.47 seconds =272.34 MB/sec
Timing buffered disk reads: 64 MB in 1.68 seconds = 38.10 MB/sec
/dev/hdg:
Timing buffer-cache reads: 128 MB in 0.48 seconds =266.67 MB/sec
Timing buffered disk reads: 64 MB in 1.66 seconds = 38.55 MB/sec
/dev/hdi:
Timing buffer-cache reads: 128 MB in 0.48 seconds =266.67 MB/sec
Timing buffered disk reads: 64 MB in 23.07 seconds = 2.77 MB/sec
/dev/hdk:
Timing buffer-cache reads: 128 MB in 0.49 seconds =261.22 MB/sec
Timing buffered disk reads: 64 MB in 24.30 seconds = 2.63 MB/sec
Is there any chance to get the 2nd controller running like the first?
Hopefully y'rs,
Hans-Peter
