20 Bře
2013
20 Bře
'13
23:35
Hi Marek,
On Mar 11, 2013, at 15:33 , Marek Vavruša wrote:
On 9 March 2013 11:51, Anand Buddhdev
<anandb(a)ripe.net> wrote:
I'm also very interested in this.
In a recent test I loaded a three large zones (one very large) into a BIND9 server with
16GB memory w/o problems. Load time about 5 min and memory usage about 13GB. NSD3 hit out
of memory, but succeeded with 32GB memory (load time around 20 min). Knot 1.2rc3 hit out
of memory also with 32GB so I went for 64GB which succeded. However, Knot took even longer
to load than NSD (and as I only have three zones, it is not the thundering herd situation)
and the final memory usage was close to 55GB!
That's a lot more memory than BIND9 and also a lot longer load time. Once loaded
everything is fine, but the memory consumption is really a concern at the moment.
Regards,
Johan
On a server with 16 GB of RAM, my instance of
BIND can load my 5174
zones into memory and use around 13 GB.
Knot didn't do so well. At some point while trying to XFR-in these
zones, it hit the memory limit and the Linux out-of-memory killer came
along and killed it.
When I started it up again, it began loading zones in from the disk, but
then appeared to go into some kind of loop, and the CPU usage was 100%.
This is usually a sign that it is stuck in some kind of loop or
deadlock. The only want to stop it is with a KILL signal (TERM doesn't
work). The log didn't output anything.
I believe this is worked on in another thread.
How can I help debug this?
Do you have any numbers on how much RAM Knot will require given a bunch
of zones? This would allow me to estimate how much RAM I will need in a
server for the zones I have.
We estimate the Knot will take around 4-5x the size of the zone on the disk,
but there are other variables and it depends on the content of the zone as well.
Also, the memory requirements may be double that during transfers, I
think it's described
in the manual.
We have a new parser that would make compiling obsolete and some
memory optimizations that would make it much leaner in preparations for the next release,
so that will hopefully help.