This little exercise involved a lot of behind-the-scenes head-scratching and consultation with Don Capps (Mr. iozone) as I tried to make sense of the results. I kept thinking that something else (besides drive performance) was causing all the read results to crash up against the 67 MB/s "Wall"—especially when I saw the VelociRaptor results!
And actually, it appears that there were OS-related issues holding back read performance, because I was able to gain around 10 MB/s of speed by switching iozone from running on XP to Vista. But that gain wasn't very stable, since moving to Vista SP1 knocked performance right back down again.
In the end, some of the main lessons learned are pretty much the same as from Part 2:
1) If you're using only one drive in your NAS, its performance is going to ultimately determine your maximum read performance. And for today's 7200 RPM SATA drives, that's going to be around 70 MB/s.
2) OS caching is the main determinant of maximum write performance, until you exceed available memory. Then drive transfer rate will determine your performance.
But there were some new revelations, too:
3) Spending big bucks on a high performance drive for a NAS may not yield the improved performance seen in internal drive applications. There are many more factors involved in network file system and copy performance that can swamp out the advantages of a "hot" drive.
4) Using Vista on a client machine can significantly improve file transfer performance. But, then again, it might not. And SP1 isn't guaranteed to provide an additional performance boost, either.
There have been some good discussions going on over in the Forums about the series so far. If you already haven't, check out the threads for Part 1 and Part 2 and join in if you have something to contribute.
So I haven't yet found the secret to building a NAS capable of 100 MB/s write and read performance. But I'm now even more convinced that one of the keys lies in effectively interleaving multiple drives to get past single-spindle speed limitations. Next time, we'll see if we can reliably break through the single-drive performance wall by bringing RAID into play.