I would be interested in seeing some comparisons of read/write speeds using swraid.device (mirror/stripe) compared to normal disk i/o.
Anyone who does this should have at least two spare harddrives as well as working/enabled UDMA on both as otherwise there isn't much point speedwise in using RAID in the first place.
I would be interested in seeing some comparisons of read/write speeds using swraid.device (mirror/stripe) compared to normal disk i/o.
Anyone who does this should have at least two spare harddrives as well as working/enabled UDMA on both as otherwise there isn't much point speedwise in using RAID in the first place.
On an A1 you additionally have to use at least two controllers ... With a single controller software RAID can only increase the seek speed, but not the transfer speed. Unless you are using very slow HDs. A driver for a hardware RAID controller would increase the speed for RAID1 writing, RAID4 and RAID5 (reading and writing), etc., but not for RAID0 (reading and writing), RAID1 reading, etc.
For comparison: With my raid.device read speed is currently up to about 60 MB/s with a 2 HD RAID1 and about 25 MB/s with a 4 HD RAID5, using file system tests (i.e. something like diskspeed, not raw device accesses with something like scsispeed or scsibench) and 2 IDE controllers. Single disk speed through file systems is about 35 MB/s.
No, if you look at it, or a picture of it you can see it's using the SII680, which means it's just the same software RAID (in it's BIOS, which isn't used on the AmigaOne, and in the included Windows driver) like with all other SII680 based controllers, not hardware RAID.
Well it should be possible, but unless both harddrives have the exact same RDB you will have to create DOSDrivers for the mirrored partitions (make sure that LowCyl and HighCyl are the same for both partitions when you create them).
"Media ToolBox" can be used to output DOSDrivers for partitions. Just remember to modify the device (swraid.device) and unit settings. Also you may want to add the line "Activate = 1" to the DOSDriver.
The same should apply to striped mode as well, except in this case you have to multiply either LowCyl/HighCyl or BlocksPerTrack by the amount of disks in your RAID.
joerg wrote: On an A1 you additionally have to use at least two controllers ... With a single controller software RAID can only increase the seek speed, but not the transfer speed.
Why do you say this? My understanding is that the bottleneck in an IDE system is not the transfer but the HD's read/write performance. Halve the amount of data the disk has to write and performance will double. The controller is capable of 133MB/Sec, just the disks that aren't, so even on one controller surely you'll get a good speed difference in transfer speed, as effectively the bottleneck's width has been doubled...?
joerg wrote: On an A1 you additionally have to use at least two controllers ... With a single controller software RAID can only increase the seek speed, but not the transfer speed.
Why do you say this?
Because I've tested it. Unlike the author of swraid.device I have the required hardware (3 IDE controllers, 4 incl. the motherboard VIA686 one, enough HDs, etc.) for testing and I've optimized my raid.device a lot already, incl. writing special versions of diskcache.library for it.
Quote:
My understanding is that the bottleneck in an IDE system is not the transfer but the HD's read/write performance.
Unless you have very old HDs the bottleneck on the A1 is the A1 For example my HDs can read/write about 75 MB/sec. sustained - if you don't use them in an A1.
Quote:
Halve the amount of data the disk has to write and performance will double. The controller is capable of 133MB/Sec, just the disks that aren't, so even on one controller surely you'll get a good speed difference in transfer speed, as effectively the bottleneck's width has been doubled...?
The max. speed you can get on an A1 with a single PCI33 controller is about 43 MB/sec., no matter if you use a single HD or several ones at the same time. By using 2 (or more) HDs connected to 2 (or more) different controllers at the same time you can get about 72 MB/sec. (raw device I/O speed, with raid.device, diskcache and filesystem overhead it's max. 68 MB/sec. reading on a 2 HD, 2 controller RAID1, but only with extreme benchmarks, for "normal" transfers you get max. 60-65 MB/sec. With other RAID systems, for example RAID5, you get much less speed, and of course writing is slower with RAID1 as well since the data has to be written to all HDs of the RAID.).
With a single HD and putting one of the controllers in the PCI66 slot of the A1, which should theoretically work since I'm using a Voodoo3 PCI gfx card, I do get the max. HD speed, i.e. more than I get using several HDs with several PCI33 controllers, but it doesn't work reliable and I get data corruption.