TCP-group 1991
[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
kh113016 version
- To: tcp-group@ucsd.edu
- Subject: kh113016 version
- From: barry@dgbt.doc.ca (Barry McLarnon DGBT/DIP)
- Date: Fri, 18 Jan 91 09:23:42 EST
Kelvin writes:
>b) The suggestion that someone made about searching the memory queue from
>the beginning made a lot of sense: And it works! Turn on the new method
>using "mem eff y". The default is still the old method of queue lookup.
To give credit where it's due, it was Dave Perry VE3IFB who suggested this
change. I've been testing his scheme transplanted into the 0828 version
on my host/packet switch (with PI/NRS/SLIP interfaces), and it has made a
remarkable difference. With the old storage allocator scheme, my coreleft
would normally dwindle down to a few K within a matter of hours after
startup. This eliminated the possibility of getting a shell, and also
seemed to make the system more vulnerable to crashes (usually preceded by
loads of alloc fails). With the new scheme, I still have 175K coreleft
two days after startup, which is completely unprecedented! The free
memory list looks much more orderly than before, and so far no nasty side
effects have been noticed.
Barry