summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorThomas Schwinge <thomas@schwinge.name>2010-07-24 12:51:44 +0200
committerThomas Schwinge <thomas@schwinge.name>2010-07-24 12:51:44 +0200
commit883d6915a37f0266479c9b25d8c911ce1e66d315 (patch)
tree26fe1f2422c3b3e0e7d7274c596c5374c647bcac
parent6dc393369a96f1629dea8fe4687a4c2e36502b89 (diff)
hurd/libihash: IRC discussion.
-rw-r--r--hurd/libihash.mdwn16
1 files changed, 16 insertions, 0 deletions
diff --git a/hurd/libihash.mdwn b/hurd/libihash.mdwn
index 58b6368c..39692939 100644
--- a/hurd/libihash.mdwn
+++ b/hurd/libihash.mdwn
@@ -8,6 +8,20 @@ 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]]."]]"""]]
+ * IRC, unknown channel, unknown date
+
+ [[!tag open_issue_hurd]]
+
+ <neal> so, we need a new ihash implementation
+ <neal> marcusb: When 80% full, the collision rate is very high.
+ <neal> marcusb: I tested using 512mb / 4096 entries
+ <neal> marcusb: Changing the load factor to 30% resulted in my program running more than an order of magnitude faster.
+ <marcusb> yeah, it shouldn't get so full
+ <marcusb> don't we do an exponential back-off in the array size?
+ <marcusb> of course it's clear we can do much better
+ <marcusb> the ihash algo is very simple
+ <marcusb> I'm not even sure it makes much sense to have a generic library
+
* Hurd libihash
* old
@@ -33,3 +47,5 @@ is included in the section entitled
* <http://cmph.sourceforge.net/>
* <http://libhashish.sourceforge.net/>
+
+ * <http://www.azillionmonkeys.com/qed/hash.html>