Lennart Sorensen wrote:
On Mon, May 14, 2007 at 12:28:10PM -0500, Neil Gunton wrote:Thanks, that's useful. I will probably end up installing using the i2o_block (if that turns out to be possible) and then roll my own using dpt_i2o. Somehow using the "official" Adaptec driver feels better. I have no idea why it wasn't included as the "official" driver... it's been working flawlessly for me since mid-2005. Not one disk error (that I can see).Well certainly when I last looked a couple of years ago, using the dpt driver on 64bit simply wasn't an option because the code wasn't 64bit compatible. Maybe there are patches to fix it, or maybe someone actually fixed it since. Not sure.
In 2005, when I was going through the original issue, I somehow managed to get in touch with Mark Salyzyn at Adaptec. He emailed me the source code for the dpt_i2o driver, which had been tweaked to work with the 2.6 AMD64 kernel. Just a couple of weeks ago he emailed me the latest version of this, so obviously they are still maintaining it internally, if on an unofficial basis.
Mark had told me back in 2005 that the dpt_i2o version was never accepted into the official kernel because the community had apparently already decided that i2o_block was somehow better, and it wouldn't be good to have two drivers that do essentially the same thing. I always thought this was rather sad, especially given that there were always doubts about the reliability of i2o_block.
In this most recent email I got from him (weeks ago), he said that the maintainer of the i2o_block driver had removed himself from the maintenance list, so the driver's future was in doubt. That's second-hand conjecture, though.
I can forward Mark's email address (and the last version of the dpt_i2o driver he sent me) to anyone who's interested.
In a nutshell, dpt_i2o works just fine on AMD64, I've been using it 24/7 on a relatively busy website (upward of 50,000 page requests per day, MySQL, apache, 450,000 pics all on the one server, RAID10) for two years, not a peep out of the hard drives that I can see. Also, I think I remember Mark mentioning that dpt_i2o might be a little faster than i2o_block. Finally, you get familiar /dev/sda devices rather than the weird ones that i2o_block gives you (not that that really matters, but I prefer standard names whenever possible).
Thanks again, /Neil