We have a similar configuration (SX6000) and for the same purpose, store GIS
files: shape files, tifs and jpg (not databases), it's like an image file
repository.
<snip>
I have a Promise SX6000 w/ 6 WD2000 EIDE drives and 128MB cache. I flashed the BIOS to the latest and am using Debian Woody with Linux kernel v2.4.25. I use the Promise SCSI drive from the website, it compiles with out any problems. I think I got i2o working, but since this driver worked just find I went with it. The system hardware is an AMD AthlonXP 2500+ w/ 512M RAM. I have 1 RAID5 array with 5 of the disks (800G) (reiserfs) and 1 hot spare. So far so good. I would have liked to not have a hot spare, but am too paranoid the office will not call me if a disk fails. On a side note, you will hear the SX6000 start beeping if a drive has problems. If the drive has not completley calved, the card will atempt to rebuild the drive. You can use the system during this time if you don't mind the beeping. If the disk rebulds, the beeping will stop and life is good. If the drive fails to rebuld you MUST replace it (the beeping will continue). I have had a rebuid occure once due to a power falure. The array rebuld just fine, I think it takes about 8-12h at 800G-1Tb storage.
I installed sysstat or more specificaly iostat and tested the system moving a 1G file of random data. I tried going to and from the RAID array and a copy on the array. In general I got 30MB/sec reading and 11MB/sec writing. I then phoned Promise and asked what I should expect and they said I was on the mark. They also commented the SX6000 is not really a performance oriented card: I am glad they were honest.
We will probably use the array for storage. People will move the GIS data to some 75G SCSI drives to do batch processing or other work. I am also going to put a fast large SATA or EIDE disk in as a hog disk, where I can spool/buffer backups before they hit the SCSI tapes.
My only real complaint with the card is the high load averages that I get when working the array. I would say if I copy 120G from EIDE to the array my load average is 1.5-1.7 for the duration of the copy (3 hours at 11M/sec). This is if the system is doing nothing else. I know this is not all the arrays fault, after all I am using the an IDE channel. But I do get a high load everage when doing a copy/move on the array. Some times the load average can be higher. But the system seems fairly resposive.
I just read a message from somone with a 3Ware card who complaind of similar concerns. Speaking of which, I phoned the people at 3Ware to ask what I would get with an Escilade card and they were not much better. Probably about 40M/sec they said but would not be as specific as Promise.
I find reiserfs works well on the array. I have another RAID5 array that does not have the hot-spare drive (to be replaced with the 2nd one I am working on) that has not lost a byte of data. When I created the array I used the 64K stripe block size. I suppose I could go smaller, and perhaps save some space but I have very few small files. Most are 1MB or more. The people at Promise were not optimistic that I would get any better performance with a different block size: given the i/o stats I am getting. In retrospect I might have gone with a 32K stripe block size, or at least tested it. But at 12h to initialize the array, I did not have an extra day.
All in all I do find you get what you pay for. I bet an SATA 3Ware card would be much better, but I don't really want to swap out perfectly good drives. I may as well get a few more fast SCSI or SATA disk for people to do work on. Speaking of drives, I find the WD2000 drives run quite HOT. I'm probably going to add some more ventilation. If you drop the temperature of your drives by only a few degrees you stand to significantly increase their life expectancy.
J.
- To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
