Browsing Google, I went into this:
http://www.geocrawler.com/archives/3/488/2001/11/0/7198728/
I have the same sort of problem. I'm trying to finish writing a traffic
accounting tool that I first designed using hash tables. But as the
traffic it was to measure kept going over 600Mbit, I decided that we
really needed something that would compute faster, leaving more CPU
cycles for the packet processing inside the kernel.
For starters, I wrote a small NetFlow aggregator. I used md5 strings as
keys for the hash table which I stored the NetFlow structures into.
That's what I also intended for my tree. As you may have already
guessed, I implemented a cleanup routine for inactive flows. So in the
first case I used g_hash_table_foreach_remove(). The problem popped up
on large numbers of flows (I had an average of ~200k). It seemed to me
that the cleanup routine ate up all the CPU. That's why I decided to
move to a faster data structure.
Now, I tried both indexing the unused strucutures in a separate hash
table, which I would clean up at a later time and destroying the tree
nodes while iterating through it. Both times I got several errors
regarding double-frees. What puzzled me was that the warnings came up
both when I manually free()-ed the keys and elements in the tree and
when the tree destroying routine did that itself.
What do I miss ? I can't recall free()-ing the data on purpose twice.
I'd like an example of an implementation from someone who has more
experience with Glib than me.
Thanks in advance for any help provided.
----
If it's there, and you can see it, it's real.
If it's not there, and you can see it, it's virtual.
If it's there, and you can't see it, it's transparent.
If it's not there, and you can't see it, you erased it.
Attachment:
pgpVccYbaAKfO.pgp
Description: PGP signature