20010801: AMI -> LSI Logic

AMI have sold their RAID group to LSI. There don't appear to be any major changes in their product lineup at this time.

       20001028: catching up

A lot has happened, and I've written about very little of it. Sorry for all the gaps.

AMI sent me "one of everything" and all the documentation that I needed. As a consequence, we now have support for all of their adapters well in time for the 4.2 release. Work is also underway on their management tools; I'm in touch with their development group on this topic and have an initial port of the MegaManager utility already complete. Hopefully AMI will be releasing more tools in the near future.

In the meantime, thanks to Elaine Morris and Peter Jarrett at AMI for the assistance, and to all my testers for their feedback and encouragement.

       20000514: all quiet...

Well, after the NDA issue was resolved, AMI have again gone quiet. Some comments from other developers working with them suggests that this is fairly typical; I guess I need to wait for another marketting emergeny. I did receive one question about their FlexRAID 'virtual sizing' feature; I tried it out here on an Elite 1500 (Dell PERC 2/DC, actually) that I currently have on loan, and it works as expected.

The 'command wedged' issues are all resolved; you can't pile more than about 126 commands into the controller, even if it claims that you can - it's not clear what happens to the commands, but they never come back out again. At this point in time, the driver in both -current and -stable is believed to be entirely stable.

I'm still hoping that someone will come up with an Enterprise 1600 for me to test on; I get the impression that I need to make some minor changes to support it.

       20000421: waiting for...

I exchanged some mail with AMI again last week; they were having trouble getting through to me, and seemed to never have received the NDA. Things are getting quiet on that front again though.

It looks like we might still have some problems though - several reports implicate the driver in system crashes. Suspicion centers on the wedged-command detection code, which completes a command as "wedged" if it's been outstanding for more than 30 seconds. This is actually rather dangerous, as the controller still has the buffer addresses and if it's just running late, it's going to kill something else. Perhaps the timeout is too short; or since it uses wall time, perhaps the clock is moving around...