[ih] early networking: error checking of main memory

[Steve Crocker]

I believe the Harvard IMP had a memory failure that resulted in it advertising that its distance to all nodes was zero.  Instant popularity :)

Host level checksums would not have caught that particular problem, but other checks would have.

Steve

Sent from my iPhone

> On Apr 22, 2024, at 4:52 AM, John Gilmore via Internet-history <internet-history at elists.isoc.org> wrote:
> 
> vinton cerf via Internet-history <internet-history at elists.isoc.org> wrote:
>> The memory of the IMPs was not error checked in the way current day
>> computers do.
> 
> At Sun in the 1980s, we put parity checks on all of our memory, and in
> later models, offered ECC main memory.  I was horrified when I moved
> from Suns to PC's in the '90s and discovered that the vast majority of
> IBM PC clones (up to this day) are shipped with no error checking and no
> error correction on main memory.  Just like those 1970s IMPs!
> 
> Some of this comes from Intel's mendacious attitude that anybody who
> wants error checking should pay them more for a CPU chip -- so they
> provide none in their cheaper CPUs, which are of course the ones that
> are shipped in the highest volumes.
> 
> On AMD chips and motherboards, both parity and full ECC have been
> supported for generations.  All you have to do to enable it is to use
> SIMM or DIMM memories that are a few bits wider (and thus a few dollars
> more expensive).  But the Lenovo laptop I'm typing on has no parity nor
> ECC on its memory, despite its AMD Ryzen processor.  (The Linux kernel's
> boot messages will include something like "EDAC amd64: Node 0: DRAM ECC
> enabled" if it's there and working.)
> 
>    John
>    
> --
> Internet-history mailing list
> Internet-history at elists.isoc.org
> https://elists.isoc.org/mailman/listinfo/internet-history