Messages in this thread | | | Date | Sat, 12 Dec 1998 16:48:42 -0500 (EST) | From | Eric Lee Green <> | Subject | Re: Dumb question: Which is "better" SCSI or IDE disks? |
| |
On Mon, 7 Dec 1998, C S Hendrix wrote:
> Regarding DPT controllers: really, they aren't that great. I like > them, and a lot of people told me before I got it that it was the > best thing since sliced bread. Easy to program, and very fast. > But I've been somewhat disappointed in the speed. Granted I only
What kind of speed are you getting? What dos "hdparm -t" say when you do it on your DPT?
On an ICP-Vortex GDT 6517 (low end 1-channel RAID controller) with 64mb cache, and three Seagate Barracuda UW 9gb configured as a 3-drive RAID5, getting about 20mb/sec. With write cache turned on my writes are pretty quick, about 12mb/sec. With write cache turned off it's a dog :-(. All of this under 2.1.131 kernel, BTW. Also, as long as you have a UPS it's safe to run it with the write cache turned on, because the write cache is automatically flushed when you shut down (at least with the version of the driver that's in the 2.1.131 kernel).
With the U2 version of that RAID controller and three Seagate Barracuda LVD 9gb, I got 24mb/sec on hdparm -t. I'm unclear why that would be so, unless they've improved the hardware somewhat since they designed the 6517, because with 3 drives LVD should not make a difference. But it did. Shrug. (Hmm, that was on a PII-450 also, while the 6517 is on a PII-350, that may have made a small difference too... same amount of memory though).
The fastest I've seen with the ICP-Vortex was a 3-channel LVD model driving six 18gb Cheetahs (5 as RAID5, one hot-spare). It maxed out at close to 26mb/sec. on that little test.
Yes, I know that hdparm -t is kind of bogus as a benchmark and doesn't reflect actual use. Running an "sqlbench" is a much better benchmark of I/O throughput. On the other hand, I find that hdparm -t tends to track sqlbench (i.e., faster hdparm -t == more performance on sqlbench), presumably because faster streaming throughput usually is associated with everything else being faster too. On the other hand, they probably don't compare well across disk subsystems (e.g. IDE drives may do well on streaming throughput, but fall over bad on 'sqlbench').
-- Eric Lee Green eric@linux-hw.com http://www.linux-hw.com/~eric "Linux represents a best-of-breed UNIX, that is trusted in mission critical applications..." -- internal Microsoft memo
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |