K
Keith R. Williams
Not flash, of course, considering the write-cycles-to-wear-out problem in a
caching application ;-)
It was ultra-low power Sony static rams, kept alive with an on-board 3v coin
cell, all mounted on a PCI card and plugged into an PCI slot. If a system
totally died and had to be replaced, you could literally pull the card from
the failed box and stick it in the replacement, and the file system manager
would scarf any unwritten data and commit it to the disks. With the coin cell,
data hold time was measured in multiple weeks, even at the highest capacity
configuration.
Neat! I wouldn't have expected SRAM to be powered for any significant time by
a coin cell. IME batteries are not well liked in mainframe applications
though. I did manage to get a few SLAC 'D'-size cells in one application
though. ;-)
Which led to an unintended application: our bios and firmware coders would use
the card to migrate their code between systems (systems that couldn't boot to
any io device in their current state), and system reliability testers used
hundreds of these cards to beat the snot out of their IO systems (think of a
TurboLaser maxed out at 192 PCI slots, and how do you do worst-case testing?)
because any single instance could consume the peak throughput of any bus it
was plugged into.
And it all started as a "midnite project"
The best ones are (at least the most fun ones . Midnight engineering and
midnight requisition were my trademarks in the '70s and '80s. ;-)