summaryrefslogtreecommitdiff
path: root/open_issues/multiprocessing.mdwn
blob: eaaa2289b9aa82e19b313d379c96dc8edf732109 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
[[!meta copyright="Copyright © 2010, 2011, 2013 Free Software Foundation,
Inc."]]

[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
id="license" text="Permission is granted to copy, distribute and/or modify this
document under the terms of the GNU Free Documentation License, Version 1.2 or
any later version published by the Free Software Foundation; with no Invariant
Sections, no Front-Cover Texts, and no Back-Cover Texts.  A copy of the license
is included in the section entitled [[GNU Free Documentation
License|/fdl]]."]]"""]]

[[!tag open_issue_documentation open_issue_hurd]]

We would expect that fine-grained, compartmentalized systems, that is,
microkernel-based multi-server systems in particular, would be ideal candidates
for applying multiprocessing.  That is, however, only true from a first and
inexperienced point of view: there are many difficulties.


# IRC, freenode, #hurd, August / September 2010

    <marcusb> silver_hook: because multi-server systems depend on inter-process
      communication, and inter-process communication is many times more
      expensive across cpus
    <marcusb> silver_hook: so you either force interrelated work on the same
      cpu, or suffer heavy penalties.  and in a typical fine-grained object
      system, all objects are interconnected!
    <marcusb> silver_hook: resources in today's systems, even in a single node
      with one cpu, but more so in a network, are very non-uniform.  scheduling
      these resources efficiently is a huge problem.  restricting the resource
      distribution policies in the way microkernel systems tend to do is posing
      serious research challenges


# IRC, freenode, #hurd, 2011-07-26

    < braunr> 12:03 < CTKArcher> and does the hurd take more advantages in a
      multicore architecture than linux ?
    < braunr> CTKArcher: short answer: no
    < CTKArcher> it's easier to imagine one server pro core than the linux
      kernel divided to be executed on multiple cores
    < braunr> CTKArcher: this approach is less efficient
    < braunr> CTKArcher: threads carry state, both explicit and implicit (like
      cache data)
    < braunr> CTKArcher: switching to another core means resetting and
      refetching this state
    < braunr> it's expensive and there is no gain obtained by doing this
    < braunr> thread migration (having a thread from a client also run in
      servers when making synchronous RPC, even handling its own page faults)
      was implemented in mach4 and is imo a very good thing we should have
    < braunr> CTKArcher: and concerning linux, it's actually very scalable
    < braunr> it's already like if all client threads run in servers (the
      kernel is the servers there)
    < braunr> rcu is used a lot
    < braunr> thread migration already takes into account smt, cores, and numa
    < braunr> it's hard to do something better
    < braunr> (here, thread migration means being dispatched on another cpu)


# debian-hurd list

On Thu, Jan 02, 2003 at 05:40:00PM -0800, Thomas Bushnell, BSG wrote:
> Georg Lehner writes:
>
> > - One promise of the microkernel architecture is better performance on
> >   multiprocessor systems, or multicomputer systems. What is the status
> >   of Gnu Mach with respect to these.
>
> This may or may not be true.  The Hurd is built around a microkernel
> architecture because of its conceptual elegance and flexibility.
> Other touted advantages may be more illusory than real, at least, they
> aren't something *we* are proclaiming is our motivation.


---

See also: [[multithreading]].