Ok, I don't want us to get us distracted, but my mind likes to spin on
cool stuff we could prototype with OGD1 or just general open hardware
projects, especially when I've had a lot of coffee.  (Hey, we should
play drinking games, but with shots of espresso instead of whiskey. :)

Hardware RAID is a mixed bag of performance and reliability.  Recently
slashdot linked to an article where some hardware and
hardware-assisted RAID cards were compared.  For the most part they
only talked about performance.  I didn't read the whole thing, but I
didn't see mention of the reliability issue.  And what I mean by that
is not the general RAID redundancy.  I have friends who have had some
bad experiences with hardware RAID1 and RAID5 getting corrupted,
despite the disk redundancy.  That is, the disks are fine, but
somehow, the data got messed up on the way there.  Either they have
bugs, or they're susceptible to EM interference (which is a bug,
really).  The on-board RAIDs seem to be the worst, with the add-on
cards being not generally a whole lot better.

Of the ATA-based RAID controllers, 3ware seems to be the least
unreliable.  Some people claim to have had problems now and then, but
for the most part, they seem to have a very low failure rate.  Also,
3ware have been especially good about releasing Linux drivers, so I
don't want to hurt their business.  I'd feel bad about that.

But in the interest of science, I think a completely open design could
have some promise.  Using FPGAs rather than microcontrollers, we could
benefit from some massive parallelism.  We could design for
reliability with some redundancy in the design.  We could throw a lot
of memory at it for huge caches (and have a configurable water-mark
for how much un-written write data is allowed to be outstanding when
giving reads higher priority).  Well, I'm getting too detailed.  But
the point is that an open design can be widely tested, and if the end
product is FPGA-based, we can do hardware revision updates in the
field.  We can also get a wide community of people to test the heck
out of it, not to mention the "many eyes" on the code, looking for
problems and opportunities for enhancement.

Perhaps we could work out an arrangement where we design and build it,
but 3ware markets and sells it.  I like cooperation.  :)

[Note:  I don't think the ECP2-50 has any serdes, so we may not be
able to use OGD1 to prototype an SATA design, but that doesn't need to
stop us.]
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to