I finally got around to installing the 3ware raid card, with 4 x 1.5TB drives, which I put in RAID5 (64KB).
As soon as I start Windows (running Windows 7 64bit) Im getting 80-100MB/s according to HD Tune, which is worse than my ICH10 RAID5 array on the mobo. But the situation gets worse over time, when the write drops to 0.1MB/s. And it copies a file at 1-2MB/s at this point.
Running torrents to the array directly locks up the client in about 15 mins of downloading.... the memory usage goes up to 1.8GB and it freezes. So the write basically drops to nothing.
I know this is old now but for anyone else the reason your getting slow writes on this controler is the "Optional BBU Modual" (Battery Backup Unit) IS NOT OPTIONAL! :-( Without it you can't safely turn on Write Cache and this is why your write speed is that of a 1900's PC! ;-)
This means that this affordable card is NOT so nicely priced! I have the same problem... Got one at the top of my price range after the good reviews only to find out the BBU was sent out to the reviewers with the damn card! 3ware you SUCK! :-P
I have looked at the prices and they are £80-100 in the UK (well from the US or Europe as the cards are easy to get here but not the BBU!) :-|
Hope this helps someone who hasn't yet got it! ;-)
You did not enable writeback; with write-through the RAID5 performance will be under 10MB/s always (with mechanical HDDs at least).
Also, using HDTune to test a RAID array will not yield correct results. You need to use something that tests on the filesystem; not artificial raw read. ATTO, CrystalDiskMark, AS SSD, Sisoftware Sandra and HDTune Pro's File benchmark do this; but the regular HDTune or HDTach is not recommended for RAIDs as you can get incorrect results.
The 100MB/s read sound like only one disk is read from at any given time; lack of read-ahead causes this.
I just got the same RAID cared (3Ware 9650SE) and 4 x 1.5TB WD Caviar Black Drives. I'm debating using RAID 10, or RAID5. With RAID5 I get more space. But I'm wondering if RAID10 will be faster overall.
Random writes on RAID10 are faster than on RAID5 on virtually all controllers, but good controllers can be very fast on RAID5, especially for reads since you can use all disks, while most RAID1+0s will only use half the drives when reading. It is possible to gear benefits out of the mirroring, but many implementations do not allow this.
Not sure about 3ware; you may actually have a chance of them having a reasonably good implementation. You should do some benchmarks and find out what works best for your workload/situation.