||[Jun. 20th, 2002|11:46 pm]
sleep cycle still screwed from last night.
and tonight we get to do it all again.
skipped dinner tonight to sleep.
woke up confused and went to the gym with nick.
feels like morning and night at the same time.
going camping tomorrow with nick, julie, and her roommate.
campground has a frisbee golf course. score.
anything to kill time until i get my house.
allergies are going to eat me alive, though.
90 minutes until downtime...
eat, clean room, make a plan.
gonna sleep most of the day tomorrow, probably.
email? hah. got a billion.
I want to write/find a Linux block device driver that sits on another block device driver and replays writes to 0 or more other hosts, reliably. Actually... not sure how hard doing network stuff in a block driver would be, though, so what would be easier is to do something like how dmesg is done... writing updates into a rolling memory buffer, and block if full, with a userspace process doing the network sending and advancing the beginning point in the buffer.
Then, the userspace daemon also accepts queries for data in given ranges from that block device from the clients... so you could have hosts rebuilding a block device from a master in real-time, at a given speed. Most of the work would be in the client (also user space) that kept a coverage map of what it knew was valid. (any recent writes it got from the master, or areas it was slowly querying...)
This is how I always wished MySQL replication worked, though I see advantages in the way they do it, too. Still, I've wanted something like the above for other things, too.
Too many projects. I should just pay somebody to write the above for me.