On 4/19/05, Richard Fish <[EMAIL PROTECTED]> wrote: > Mark Knecht wrote: > > >I need to learn the RAID levels, but what I mean is I think what's > >called mirroring. > > > > Yep, thats RAID1. Forgive me, I've been married to my laptop for too > long, and I forget that 'normal' computers can have multiple 1394/USB > controllers! In that configuration, yes, you should be able to get some > decent bandwidth. > > If/when you do try this, please report the results. I am using USB2.0 > disks for my backups right now, but my bandwidth is limited to 20MB/sec > total. Since one of the disks I backup is also a USB2.0 disk, my > effective bandwidth is about 10MB/sec for much of my backup (transfer > from USB2.0 disk to memory, memory to other USB disk). > > I also have a 1394 port that I am not using for anything right now, and > I know that if I moved one or the other to that bus, I could get some > more speed. But I am not sure which one to move, because I don't know > whether it is faster or slower than the 20MB/sec maximum that I get now.
On my 3 1394 drives I get about 18MB/S, 24MB/S and 24MB/S. The 18MB/S drive is the oldest (and smallest at 40GB) of the three. 1394 bandwidth is currently limited by lack of gap count optimization in the drivers. Just playing around I've been able to get about 35MB/S from one of my drives but the problem is the settings won't tick and it eventually drops back down to the numbers above. If you were interested in trying it there is a small programm called 1394commander that has a few commands that allow setting gap count and getting faster settings, at least until a 1394 bus reset comes along. However for doing backups it may be worth it. > > >Two or more drives with identical data for > >redundancy. In my case I have three 1394 controllers in the same > >machine. I was considering putting identical drives on each cable in > >parallel so the 1394 bandwidth is essentically trippled and the same > >data is written and read to all drives. Seems to me the only overhead > >is then 3x disk bandwidth across the PCI bus as well as the > >verification that all 3 drives return the same data. > > > > > > I am pretty sure that when reading data, the kernel's software RAID > treats the the array the same as RAID0..that is it would read all 3 > disks simultaneously, but different blocks from each, to maximize > throughput. If the array is 'clean', and all 3 disks contain the same > data, there is no point in reading the same data from all 3 drives. OK, so I'll bite. If you don't actually read all three disks then how do you know they all contain the same data? I'm concerned that there could be a time when data is written and it's corrupted due to problems in the 1394 sub system. I assume for now that these corruptions are random and do not happen on all three cables. How do I guard against that? I know the drives themselves are good as I've used them all under Windows for quite a long time with no lost data ever. (Yea FAT32!) ;-) However I have lost data once already under Linux using ext3. The system said it was moving a 1GB file, however it finsihed with no error messages in 1 second and while there was a file icon on the target drive the system said the file was corrupted and couldn't read it. > > Write performance suffers slightly with RAID1, because you basically > have to wait for the slowest disk. Even if they are all the same speed, > there is always one that lags a few ms because the platters are not > synchronized. Makes sense. > > For the filesystem, I say choose your favorite. Just remember that 3 > copies of the same mistake doesn't help you much...you still need backups! Yep! Always. Thanks, Mark -- gentoo-user@gentoo.org mailing list
