Discussion Notes -- Fuzzballs

Questions

1.  What if the user doesn't respond immediately?  For example, if
they've asked for a two hour block but the machine is free in the
middle of the night, what happens?  Suppose they don't get to it until
late the next afternoon?

<don't know.  one idea is to include this in the time the user
requested, and if they go too far past their indicated time, start
reminding them and notifying "system admin".

One suggestion made in the meeting was that we should have different
"time slice" machines, so that people sign up for the one hour
machine, the 24 hour machine, etc.

Another notion was to make the amount of time that the system would
let a machine "wait" for the user depend on how many are in the queue.
I.e., lots of people waiting, very little "lag" in the system.  No one
waiting--take your time.

[NOTE: may want to make the user "notify" the server when they start
using the machine--this will allow better control of the "lag" between
notification and starting use.]

3.  Do the users have to go to the machines?

at this point in our rough design, no.  they would use telnet etc. to
access the cpu over the net.  (might use screen--allows user to
"leave" without logging off?)

So the machines could be racked somewhere convenient, just as long as
there is power and drops.

4.  What is the provision for a professor to reserve a block of
machines at a specific time?

5.  How are we going to arrange for maintenance of the machines?

6.  How do you prevent people from using them as dialups?  What
is to prevent someone from using this as a single user dialup?

[I'm confused--in essence, this IS a single user dialup.]

7.  What if a machine panics?

8.  What if a user isn't available when their machine becomes
available?  (this seems to be the main thread--I envisioned letting
machines sit idle, but perhaps that isn't acceptable)

9.  Scaling issue?


-----------------------------------------------------------

server		user		machine
	<------ I'm done
	----------------------->cleanup
				(if NG, reinstall?)
	<---------------------- okay, ready for use
	-----------------------> put next user in password file
	<---------------------- okay
	-------> machine ready
	<------  got it!
			-------> telnet, screen, etc. access
				usage!
	<------ I'm done

is the gap between 
	-------> machine ready
	<------  got it!
controlled, or simply part of the user time?

One model would be to let the server send a message, and the user has
the machine until they send "I'm done" message.

We can warn of overrun, and even force cleanout at N + X time.

I.e.
1.  Set up machine
    let user know it is available
    at N + something - alert supervisor?  alert user?
	block access?  blow them away?

2.  Alert user that machine is available
    when user responds, change password and handout machine
    (if too late, user goes back in queue and machine gets handed to
	next in line)

Suggestion: make small parts, and use a "rules" file at server to
control sequences, parameters, etc.

I.e., the business rules of this system are still pretty fluid.
Provide for that.