1 files changed, 66 insertions, 0 deletions
diff --git a/open_issues/mach_federations.mdwn b/open_issues/mach_federations.mdwn
new file mode 100644
index 00000000..50c939c3
--- /dev/null
+++ b/open_issues/mach_federations.mdwn
@@ -0,0 +1,66 @@
+[[!meta copyright="Copyright © 2012 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]]
+
+
+# IRC, freenode, #hurd, 2012-08-18
+
+    <braunr> well replacing parts of it is possible on the hurd, but for core
+      servers it's limited
+    <braunr> minix has features for that
+    <braunr> this was interesting too:
+      http://static.usenix.org/event/osdi08/tech/full_papers/david/david_html/
+    <braunr> lcc: you'll always have some kind of dependency problems which are
+      hard to solve
+    <savask> braunr: One my friend asked me if it's possible to run different
+      parts of Hurd on different computers and make a cluster therefore. So, is
+      it, at least theoretically?
+    <braunr> savask: no
+    <savask> Okay, then I guessed a right answer.
+    <youpi> well, theorically it's possible, but it's not implemented
+    <braunr> well it's possible everywhere :p
+    <braunr> there are projects for that on linux
+    <braunr> but it requires serious changes in both the protocols and servers
+    <braunr> and it depends on the features you want (i assume here you want
+      e.g. process checkpointing so they can be migrated to other machines to
+      transparently balance loads)
+    <lcc> is it even theoretically possible to have a system in which core
+      servers can be modified while the system is running? hm... I will look
+      more into it. just curious.
+    <savask> lcc: Linux can be updated on the fly, without rebooting.
+    <braunr> lcc: to some degree, it is
+    <braunr> savask: the whole kernel is rebooted actually
+    <braunr> well not rebooted, but restarted
+    <braunr> there is a project that provides kernel updates through binary
+      patches
+    <braunr> ksplice
+    <savask> braunr: But it will look like everything continued running.
+    <braunr> as long as the new code expects the same data structures and other
+      implications, yes
+    <braunr> "Ksplice can handle many security updates but not changes to data
+      structures"
+    <braunr> obviously
+    <braunr> so it's good for small changes
+    <braunr> and ksplice is very specific, it's intended for security updates,
+      ad the primary users are telecommunication providers who don't want
+      downtime
+    <antrik> braunr: well, protocols and servers on Mach-based systems should
+      be ready for federations... although some Hurd protocols are not clean
+      for federations with heterogenous architectures, at least on homogenous
+      clusters it should actually work with only some extra bootstrapping code,
+      if the support existed in our Mach variant...
+    <braunr> antrik: why do you want the support in the kernel ?
+    <antrik> braunr: I didn't say I *want* federation support in the
+      kernel... in fact I agree with Shapiro that it's probably a bad idea. I
+      just said that it *should* actually work with the system design as it is
+      now :-)
+    <antrik> braunr: yes, I said that it wouldn't work on heterogenous
+      federations. if all machines use the same architecture it should work.