|
|
1.1 root 1: .SH
2: Summary
3: .PP
4: The current
5: .UX
6: kernel implementation is quite inflexible,
7: closing off what might be interesting design choices.
8: Lightweight processes both in the kernel and in user processes could
9: be used to good advantage.
10: The kernel is not properly structured to allow easy use of different
11: facilities together.
12: Streams may be a decent first step in this direction.
13: .PP
14: Stub generators, message passing and
15: RPC transport protocols all
16: need substantial work as
17: .UX
18: moves into the distributed world.
19: Using these protocols without stub generators is like a day without sunshine.
20: .PP
21: Resource location, authentication and naming are issues
22: .UX
23: has not faced in the distributed
24: environment.
25: Cascaded services present another level of issues which need to be faced
26: in their design.
27: .PP
28: .UX
29: has ascii terminals ingrained into its very nature.
30: It will take much more work to smooth the rough edges emerging from
31: the forced marriage of workstation displays with
32: .UX .
33: .PP
34: If a system resource is in short supply (as file descriptors are),
35: the correct solution is to lift the limit entirely.
36: Doubling or tripling a limit on a resource only delays the day of
37: reckoning,
38: while still preventing those design strategies that found them in
39: short supply originally.
40: .PP
41: Shared memory should allow sharing of memory between processes,
42: between the kernel and a process, and between a process and hardware.
43: Shared libraries would open up design opportunities.
44: .PP
45: More work needs to be done on performance of some of the new kernel
46: facilities.
47: The X server uses select more heavily than any other system call,
48: accounting for the largest single component of CPU time used,
49: though select is not the limit in absolute performance.
50:
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.