Hardware for EROS?

[Mark S. Miller]

At 12:13 PM 4/11/00 , Jonathan S. Shapiro wrote:
>I think in practice that the checkpoint really has to be stabilized to
>non-volatile storage. If there is any practical chance that the battery dies
>on you, the entire system ends up unrecoverable unless the checkpoint has
>also been stabilized.

How much less reliable is battery-backed-up ram than a disk?  Since we make
disks reliable with redundancy (mirroring & raid), why not do this for
battery-backed-up ram?

Also, it's properly a policy decision by the proprietor of a given machine
whether it's worth risking recoverability on non redundant batteries in
exchange for low latency commits.  For many things I'd like to use EROS for,
I'd take this chance.

Remember, machines are cheap.  It's not unreasonable to partition
computation onto machines according to their need for reliability vs low
latency commits.


         Cheers,
         --MarkM