Tests with 3Ware's 6410 ATA-raid controller on Linux 2.4.x

for full details check the 2400a page.

Note, 8.10.02: The 6410 has been discontinued. Please check 3ware's webside for their current products. They probably do ship newer software versions by now too. Experience with them may be better or worse.

Raid BIOS

On init of a new RAID-1 (mirror) array, 3ware's BIOS does not allow copying of one of the drives to the other, both drives get written over. While "profiling" both drives, we're stuck in the 3ware BIOS screen. What did I buy a RAID with an extra on board controller for, 3ware folks ?

Kernel

hardware
K6-500, 128MB
software
vanilla 2.4.16 SMP kernel compiled from scratch, 3ware: firmware FE6X 1.02.28.053 (Jan 2002)
The standard 2.4.x kernel recognizes the 6410 when compiled with 3ware RAID (in low level SCSI drivers):

3ware Storage Controller device driver for Linux v1.02.00.016.
PCI: Found IRQ 11 for device 00:08.0
scsi1 : Found a 3ware Storage Controller at 0xe000, IRQ: 11, P-chip: 5.7
scsi1 : 3ware Storage Controller
Vendor: 3ware Model: 3w-xxxx Rev: 1.0
Type: Direct-Access ANSI SCSI revision: 00
Attached scsi disk sda at scsi1, channel 0, id 0, lun 0
SCSI device sda: 78240928 512-byte hdwr sectors (40059 MB)
sda: unknown partition table
SCSI device sda: 78240928 512-byte hdwr sectors (40059 MB)
sda:
3w-xxxx: scsi1: AEN: Initialization started: Unit #0.

Notes:

  1. a standard (non-raid) IDE disk appears as a standard SCSI disk and is thereby accessible thru the 3ware controller
  2. the kernel, lilo, mount etc. work as usual and don't get disturbed by the underlying RAID nature

User space

To access the raid statistics and management, a webserver 3dmd allows monitoring and configuration. Here're three config screens:
Links on the 3dmd webpages below obviously don't work, as they are not the accessing the live system.
home details configure
Running a webserver for configuration of the RAID opens a potential risk for outside interference. 3dmd doesn't filter for IP numbers, doesn't do https, but allows a password. For remote administration, a filtered https/http rewrite with an Apache server would one way to enhance security.

converting back to std IDE

Out of the RAID-1 (mirror) array (partitioned and with ext2 filesystems), one disk was removed and attached to standard IDE controller. fdisk fails to recognize the partition table. So there's no way back should the 6410 itself fail. If you're really serious, go buy an identical one right from the start and shelve it.

Performance

Test scenario:
Two 40GB 5400rpm IDE ATA100 disks: Samsung SV4002H (2MB cache) and a WDC WD450AA (2MB cache) as Raid-1 (mirroring).
c't magazine measured minimal continous transfer rates [MB/s] of WD450AB (15.2) and SV4002H (15.7).
I didn't do the test with a graphical output like the one's with the Adaptec 2400a, here are the raw numbers (note that the motherboard and disks were slightly different, which IMHO doesn't matter):
Reading from partition (/dev/sda5) was 22MB/s, writing 20MB/s , both higher than the 2400a. Read performance dropped to 1 MB/s while a concurrent write to another partition (20MB/s) was active.
During init (while Linux bootet):
Jan 20 14:11:46 pluto kernel: 3w-xxxx: scsi1: AEN: Initialization started: Unit #0.
Jan 20 14:55:11 pluto kernel: 3w-xxxx: scsi1: AEN: Initialization complete: Unit #0.
read performance dropped to 2MB/s .

Results

  1. I/O rates for our scenario were higher (20MB/s) than that of the 2400a (13MB/s)
  2. I/O rates drop during array init (2MB/s).
  3. The BIOS lacks support for backgrounding an array init (this is called profiling by 3ware).
  4. Contents of a drive out of a RAID-1 array is not accesible with a standard IDE controller.
Please bark at info@pab-opto.de in case you feel upset or interested.
updated: