[mythtv-users] Fwd: Further Notice of Seagate Hard Drive Class Action and Proposed Settlement

[f-myth-users at media.mit.edu]

    > Date: Fri, 12 Mar 2010 16:18:54 -0800
    > From: Manuel McLure <manuel at mclure.org>

    > On more modern systems, multiple bytes (usually 8) and their parity
    > bits are combined - this is what's known as ECC. The ECC can detect
    > and correct single bit errors, and can detect two bit errors but not
    > correct them. In that case it will crash the system exactly in the
    > same way as single byte parity does.

Not just modern.  Even the Lisp Machine, whose design dates from the
70's and 80's (depending on which one) used SECDED, and it wasn't even
some big-iron mainframe.  It correctly notified the user about all
recoverable and unrecoverable RAM errors, and greatly contributed to
avoiding weird crashes and flaky memory.  [In the case of the Ivory,
for exactly, a memory word was 40 bits (32 data + 8 tag), plus 8 more
bits of SECDED, hence a 48-bit-wide datapath.  (The number of SECDED
bits is logarithmic in the word width, so while the correction
hardware gets expoentially more complex, the RAM overhead of
correction bits goes down rapidly as the words get wider.)]

This was from an era when RAM was considered unreliable, but you what?
It's- still- unreliable in many cases.  Witness the sheer number of
memory diagnostics and the "check you RAM!" mantra whenever something
odd starts happening.  It's only cheap commidity hardware that decided
to just jump into the weeds instead of checking for errors in the
first place.