summaryrefslogtreecommitdiff
path: root/doc
diff options
context:
space:
mode:
authorJoey Hess <joey@kodama.kitenet.net>2008-11-11 13:40:02 -0500
committerJoey Hess <joey@kodama.kitenet.net>2008-11-11 13:40:02 -0500
commit700c4bef294a792d507f6cfc34dd4dd5427f2a7f (patch)
treee4ea377a3103b1715eab59ef4a2f625f0c6ae9b4 /doc
parent8c47f16dccd8a4dd91f365233690bd8139431249 (diff)
let's stop sucking :-)
Diffstat (limited to 'doc')
-rw-r--r--doc/todo/avoid_thrashing.mdwn20
1 files changed, 20 insertions, 0 deletions
diff --git a/doc/todo/avoid_thrashing.mdwn b/doc/todo/avoid_thrashing.mdwn
new file mode 100644
index 000000000..6c895e7c9
--- /dev/null
+++ b/doc/todo/avoid_thrashing.mdwn
@@ -0,0 +1,20 @@
+Problem: Suppose a server has 256 mb ram. Each ikiwiki process needs about
+15 mb, before it's loaded the index. (And maybe 25 after, but only one such
+process runs at any time). That allows for about 16 ikiwiki processes to
+run concurrently on a server, before it starts to swap. Of course, anything
+else that runs on the server and eats memory will affect this.
+
+One could just set `MaxClients 16` in the apache config, but then it's also
+limited to 16 clients serving static pages, which is silly. Also, 16 is
+optimistic -- 8 might be a saner choice. And then, what if something on the
+server decides to eat a lot of memory? Ikiwiki can again overflow memory
+and thrash.
+
+It occurred to me that the ikiwiki cgi wrapper could instead do locking of
+its own (say of `.ikiwiki/cgilock`). The wrapper only needs a few kb to
+run, and it starts *fast*. So hundreds could be running waiting for a lock
+with no ill effects. Crank `MaxClients` up to 256? No problem..
+
+And there's no real reason to allow more than one ikiwiki cgi to run at a
+time. Since almost all uses of the CGI lock the index, only one can really
+be doing anything at a time. --[[Joey]]