Posted by
CmdrTaco
on from the now-thats-a-lotta-pr0n dept.
Delta-9 writes "Here is a writeup on how to combine 6 200GB IDE drives into a small tower and hack together some firewire controllers to give you one giant 1.2TB firewire drive." Very cool project, both technically and aesthetically.
Forgive my hardware ignorance but...
by
Kethinov
·
· Score: 4, Interesting
if you use firewire controllers to give you one giant 1.2TB firewire drive, doesn't that essentially make 6 hard drives pretend to be one? (AKA the OS doesn't realize it's many) And if just ONE of those drives failed, aren't you shit out of luck with your data?
Again, forgive my hardware ignorance if I'm way off.
-- You're right, I wouldn't steal a car. But if it were possible, I sure as hell would download one!
I tried this on Linux with some problems.
by
Rolman
·
· Score: 4, Interesting
I tried this on Linux and got terrible performance at the first try, I got a 23MB/s RAID-0 when each HDD is capable of 26MB/s by itself (everything according to Bonnie++ and hdparm). I didn't know what to blame, the bus, the cables, the Linux SCSI layer, or the whole IEEE1394 support on Linux. Windows was noticeably faster with up to 28MB/s.
Then I made some more research and it turned out the problem was caused by the sbp2 kernel module. This module had some good fine-tuning parameters (sbp2_max_sectors, sbp2_max_outstanding_cmds and spb2_max_cmds_per_lun) up to 2.4.20, but these got ditched in 2.4.21 in the name of a "better way of handling these parameters". I understand the logic behind this move, but the tweakable granularity should have been kept.
Using 2.4.20, I managed to get better performance by tweaking these parameters, then modified sbp2.c on 2.4.22 to reflect the changes. However, I haven't been able to get the 35MB/s this guy got so easily on MacOS X, I'm currently stuck at 29.22MB/s maximum and it's painfully slow to test all combinations of those variable parameters on the sbp2 module.
I just wish there was some document which could explain more about the relationship between these parameters for people not actually involved on the linux1394 project. The comments on sbp2.c are not helpful beyond this point.
By the way, I'm using two Oxford-based bridges to connect two 8MB cache Matrox HDDs, and I'm using Bonnie++ and hdparm for testing. YMMV but the least I can say is Linux RAID support on Firewire still has a long way to go.
Slashdot Mirror
if you use firewire controllers to give you one giant 1.2TB firewire drive, doesn't that essentially make 6 hard drives pretend to be one? (AKA the OS doesn't realize it's many) And if just ONE of those drives failed, aren't you shit out of luck with your data?
Again, forgive my hardware ignorance if I'm way off.
You're right, I wouldn't steal a car. But if it were possible, I sure as hell would download one!
I tried this on Linux and got terrible performance at the first try, I got a 23MB/s RAID-0 when each HDD is capable of 26MB/s by itself (everything according to Bonnie++ and hdparm). I didn't know what to blame, the bus, the cables, the Linux SCSI layer, or the whole IEEE1394 support on Linux. Windows was noticeably faster with up to 28MB/s.
Then I made some more research and it turned out the problem was caused by the sbp2 kernel module. This module had some good fine-tuning parameters (sbp2_max_sectors, sbp2_max_outstanding_cmds and spb2_max_cmds_per_lun) up to 2.4.20, but these got ditched in 2.4.21 in the name of a "better way of handling these parameters". I understand the logic behind this move, but the tweakable granularity should have been kept.
Using 2.4.20, I managed to get better performance by tweaking these parameters, then modified sbp2.c on 2.4.22 to reflect the changes. However, I haven't been able to get the 35MB/s this guy got so easily on MacOS X, I'm currently stuck at 29.22MB/s maximum and it's painfully slow to test all combinations of those variable parameters on the sbp2 module.
I just wish there was some document which could explain more about the relationship between these parameters for people not actually involved on the linux1394 project. The comments on sbp2.c are not helpful beyond this point.
By the way, I'm using two Oxford-based bridges to connect two 8MB cache Matrox HDDs, and I'm using Bonnie++ and hdparm for testing. YMMV but the least I can say is Linux RAID support on Firewire still has a long way to go.
- Otaku no naka no otaku, otaking da!!!