[E-Lang] FW: Reasearch ideas in HDC

Bill Frantz frantz@communities.com
Mon, 29 Jan 2001 16:58:14 -0800

At 08:49 AM 1/29/01 -0500, Ralph Hartley wrote:
>How do E and EROS handle resource leaks? Persistence can make the  
>problem much worse. In the quote it is implied that they treat the  
>problem by rebooting the computer, which frees all (most) resources. If  
>I understand correctly, in EROS this approach will not work at all,  
>rebooting leaves the system in exactly the state it was in before.
>What do you do when a system like EROS gets totally clogged by leaked  
>resources? Don't say we will keep that from ever happening - it WILL 

KeyKOS and EROS provide, thru the SpaceBank, a mechanism to forcibly
reclaim space.  This logic is similar to the mechanism in file systems to
reclaim space when files are removed/deleted/erased.  In KeyKOS, we ran
command shells in "contexts".  Each context had its own space bank, and
could be reclaimed whenever the user desired.  Normally when you were
testing new code, you created a new context to tested it.  When the test
was done, you destroyed the context.

The worst problem I found with this mechanism is knowing before hand what
you were about to destroy.  The problem is similar to trying to recover
space on a Windows system when you don't know what half the files do.

>From a more philosophical point of view, the issue is not whether you leak
resources, but how fast you leak them.  I find that frequently it is more
reasonable for me to buy a new, bigger hard disk than it is to categorize
all the files on the old one and delete the ones I no longer need.
Sometimes I even plan to allow continuously growing resource use.  My
archive of this list is an example.