Topic: Maxtor / Promise Ultra133 card ok in 6.03b

[Winston]

I was just setting up a new SME server on an old PII and wanted to use a Maxtor/Promise Ultra133 addon card to support a couple of U133 drives in a mirror configuration.  I found a "How To"  through these forums and was prepared to follow there advice but the 6.03b install recognized the card immediately and built the raid array on the 2 drives.  Very straight-forward.

[Alexander Ziemann]

Are you shure, that both HDs are used?

Other folks here reported, most of these Highpoint / Promise adapters just work as "normal" controllers (JBD-Mode in Raid-Speaking).

And IF it works ok (both disks used, Raid 1 install possible -> "Single Disk") then i would switch back to e-smith as real fileserver.

 Performance got so bad even with U2Wscsi and software raid.

az

[Reinhold]

Alexander,

Note: Winston said MIRROR like in "RAID-1 (Mirroring) Mode" ...no HW or BIOS used/needed.
(I'm of course assuming that "a couple" here means 2 ...like he says in the end .-)

My 2cents:
Since I did it several times now, I too feel pretty confident that the Promise Controllers are a "perfect" companion for a 6.03 - definitely for "< 40 $/€" !

FWIW:
As an example I have running a 6 IDE-HD Software-Raid using secondary onboard IDE + 1 Promise TX2 controller. This "all SOFTWARE" RAID 5 setup yields a measured 7,2 Mbyte/s (+- 0,2) WRITE-file speed with an SME Vers. 6.03 on a mere 1GHz AMD _across_ the served 100Mbit Network (XP & ME clients).
- What more could I possibly want ? ... i.e. for _any_ price .-)
(OK I do not run 50 websites/databases or the like on it .. but just serve files in a small net .-)

As to your "slow" U2W SCSI:
... Do you have enough "horsepower"?...
Note: With a 400MHz cpu the same setup yields only 5,6 MB/s and at 233 MHz we crawl with 90% cpu load below 4MB/s ... again all "across-100Mbit-net-using- XP" ... but that's all I look at here

[Alexander Ziemann]

Hi, Reinhold wrote:

> Note: Winston said MIRROR like in "RAID-1 (Mirroring) Mode"
> ...no HW or BIOS used/needed.

I meant RAID-1 also. But what i meant was HW-RAID like it is done with my old mylex on E-Smith or Promise TX2000 on NT 4 -> setup the RAID in BIOS or by extra-software and forget it.

> 6.03 - definitely for "< 40 $/�" !

OK, but for 10 $/¤ i get a "normal" additional HD-IDE controller here - that is no RAID, but just adding HDisks.
 
> As an example I have running a 6 IDE-HD Software-Raid using
> secondary onboard IDE + 1 Promise TX2 controller. This "all
> SOFTWARE" RAID 5 setup yields a measured 7,2 Mbyte/s (+- 0,2)

How do you set up that stuff? The normal E-Smith only supports SW-RAID 1 (mirrored disks).

> WRITE-file speed with an SME Vers. 6.03 on a mere 1GHz AMD

My SME Server at Office is a P III/800 and CPU-load is low (< 10 %) even with big filecopies.  Before that SW-RAID 1 and 2 U2W SCSI drives CPU load was up to 90 % on the same machine, when we did big filecopies or put heavy load on the server.

> Note: With a 400MHz cpu the same setup yields only 5,6 MB/s

At the server?

az

[Reinhold]

Alexander,

YES "in 2003" one simple additional channel is enough for Raid1
... it could be even less than 10 bucks if you use that "free" 2nd channel on your Mainboard.
Personally I don't use Raid1 thinking that it's better to go all the way to RAID5 if you want a large amount of data stored (economically).

SETUP & FORGET can be had in SME 6.03 ... actually it's the DEFAULT MODE .
Just plug in 2 HDs before you start the SME CD install - everything will be done "automagically" by SME 6.03 itself.
(Note: If you want hotplugging and stuff you will still need/want some "metal&more" and ...there are advantages to the Promise chipsets (like 1 IRQ only for 4 HDs)).

Raid is already in the SME 6.03 kernel. For setting up Software-Raid5 etc. you will want to look at the excellent Howto by Jakob Østergaard http://www.tldp.org/HOWTO/Software-RAID-HOWTO.html or, assuming you can read that, go for selflinux.org
In addition you might want to scan this forum for some "/etc/raidtab"s including my own (another example where one should add/fix persistent superblocks .-).

All data given were across 100Mbit net ... and you have to give me more before I understand the gist of that last question .-)

[Reinhold]

I forgot to say:
"cat /proc/mdstat" will show how many HDs are used ...
(or if you do have a raid at all)
.

... and many tools will work with /dev/md0 (the raid array)  !

[Winston]

Reinhold, Thanks for the command info.

"
Welcome to the Mitel Networks SME Server.
[root@mail root]# cat /proc/mdstat
Personalities : [raid1]
read_ahead 1024 sectors
md2 : active raid1 hde3[0] hdg3[1]
      262016 blocks [2/2] [UU]

md1 : active raid1 hde2[0] hdg2[1]
      59669952 blocks [2/2] [UU]

md0 : active raid1 hde1[0] hdg1[1]
      102208 blocks [2/2] [UU]

"
Winston

[Reinhold]

>root@mail root]# cat /proc/mdstat
(1)>Personalities : [raid1]
>read_ahead 1024 sectors
(2)>md2 : active raid1 hde3[0] hdg3[1]
(2a)>262016 blocks [2/2] [UU]
>
>md1 : active raid1 hde2[0] hdg2[1]
>59669952 blocks [2/2] [UU]
>
>md0 : active raid1 hde1[0] hdg1[1]
>102208 blocks [2/2] [UU]

Hi Winston,

*(assuming here you want me to comment to the above so I just "type away" .-)

(1)You got a working raid 1 for a basic 3 partition e-smith setup  ... albeit a bit unconventional .-)
Normally  /dev/hda is 1st HD - the masterdrive on the first internal IDE channel - however this is not used in your list (Am I right to assume you got your CD-ROM as the lonely master on the 1. IDE channel ?-)

(2,2a)You use /dev/hde   (3 partitions hde1 - hde3) i.e. the secondary master drive...
and it's "raided" with  /dev/hdg_   (3 partitions hdg1 - hdg3) on the primary-master on your new Promise card.

- Each partition on hde (1,2,3) is mirrored on hdg with the new raid1 dev name of "md0","md1" and "md2".
- All 3 md_ raid1 drives  work as shown by those three "[2/2] [UU]"
(if a drive fails you will see like "[1/2] [_U]"

...guess that's it .-)
(but have a look at the HowTo mentioned - good reading .-)

[Reinhold]

damn ... /dev/hda should read /dev/hdc ... so correct is:
"Normally /dev/hdc is 1st HD - the masterdrive "...

[Dan Brown]

Actually, I think you were right the first time; /dev/hda is the primary master hard drive.  /dev/hdb is the slave on the primary interface, hdc is the master on the secondary, and hdd is the slave on the secondary.  If you have a third and fourth channel, they'll be hde/hdf and hdg/hdh respectively.

[Winfried Wilcke]

I just finished builiding two identical servers. Both show the same behavior of  truly awful  performance with software mirroring enabled and work fine without raid.
The systems each have 1.4 GHz Athlon,  2x 120 GB Western Digital 100UDMA
drives, 256 MB RAM.

RAID-1 is turned on    /proc/mdstat output looks as one would expect,
3 partitions, mirrored on hda and hdc, and all active on both drives.  
The two IDE drives are both the masters of their channels, a CD -R/W drive is a slave on the primary IDE channel.

Performance, measured by transferring 26 files of about 14 MB total:

0.5 Mbits/s when RAID1 turned on, transferring across 100Mb/s Ethernet
1   MBytes/s  when RAID1 turned on, transferring locally within the server

36 Mbit/s with only one disk enabled, across the Ethernet
sub-second time to transfer 14 MB locally, perhaps 20 MB/s ?

Any idea where to look for fixing this performance problem ? I'm not exactly
a Linux guru!
cheers /www

[Alexander Ziemann]

Hi Winfried,

you should put away the CD-R/W; i once checked a very similar setup and the performance was ugly.

I actually now disable or take out CD-R after installation in order to free the IDE-Bus.

az

[Winfried Wilcke]

Alexander, thanks for your help. I removed the CD writer, and indeed, the performance improved by a factor of 6. It takes now 42 seconds to transfer 14 MB, rather than 4 minutes.
The absolute performance is still about 15 times slower than it should be, compared to another machine with the same software and even a slower processor. However, I realized also that  DMA is not turned on and I need to get a kernel patch to support the chipset (VIA KT333) to be able to activate DMA.
Thanks again for your help!  Winfried