>>>>> "wm" == Will Murnane <will.murn...@gmail.com> writes:
wm> http://www.newegg.com/Product/Product.aspx?Item=N82E16820148115
wm> $48 Unbuffered (intel x38 chipset takes these, for example)
yeah for some reason they only make the slow speeds in unregistered
ECC which is annoying. I think AMD has just released or is about to
release the desktop CPU's with the faster memory controller, but you
can't get faster ECC sticks for them, only fast non-ECC. Also I am
worried the ECC error is not handled by the whole software
stack---maybe some ``no, seriously, check ECC shit'' knob is not
twisted by the cheapo Chinese gamer BIOS, and it is only BIOS writers
who worry about this bit since they are supposed to enumerate all the
RAM sticks and their SROM's, not the OS. Or the CPU / northbridge
supports ECC but the pins aren't wired on the motherboard. Or
technically Phenom II checks ECC, but the driver in the OS only works
with older Opterons, so you are hitting ECC errors all the time but
when you hit one the machine just powers down, or silently continues
healing errors with no warning until the stick goes totally bad thus
behaving in effect exactly like non-ECC RAM. Or something. Also
there's the question *when* is the error detected---when the ARC grows
that big, or does the ``memory scrubber'' thread or ``background page
zeroing'' thread find it, does the error put the machine into
syslog/fmd convulsions using up all disk bandwidth with synchronous
writes to some XML file ``ECC error, detected! detected again!'',
u.s.w. I need some kind of deliberately broken ECC test stick. With
ZFS you can scribble on disks then scrub them, but with ECC on modern
CPU/motherboards it is kind of like reading the future in chicken guts
and hoping you got it right.
pgpyt5CpTASG9.pgp
Description: PGP signature
_______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
