Hi Klaus,
do I understand correctly that you are missing a tool (e.g. supplied
along with Knot DNS/Knot Resolver), which would give you how much memory
is actually used, along with number showing how much is just "cache"?
Would you be willing to use it to get numbers about memory usage instead
of whatever tool you use now?
Again, please do not forget that comparing numbers (as you say, "without
analyzing virtual vs resistent vs ... memory") does not say anything
about behavior under stress. BIND and also Knot DNS have their journal
on disk, so if the system does not have enough memory for cache both of
them will be slow(er) to read data from disk, slowing down other operations.
The only difference here is that BIND's memory consumption (at first
glance) does not include amount of memory consumed by cache, and Knot
DNS's memory consumption (again, at first glance) does include the cache.
However in both cases the system will just drop the cache if it needs
more memory, so there is non actual difference. (This difference is
caused by way how in Linux kernel memory statistics work, it is not a
choice of Knot DNS developers.)
https://symas.com/understanding-lmdb-database-file-sizes-and-memory-utiliza…
describes this in more detail.
So, would a tool printing "real" memory consumption help you?
Petr Špaček @ CZ.NIC
On 18.3.2018 14:02, Klaus Darilion wrote:
Just some comments from a user: It is hard to guess if
the consumed
memory is something to care or not. I do not know if all the virtual
memory may be used and some time may cause problems. If my monitoring
tells me "100 % memory used" I am concerned, as I do not know if this is
a real problem or just some irrelevant buffers. Also I do not know how
the name server will behave when running out of memory (crash, serving
old data, serving broken data, ...). Hence, I have to add more memory or
choose a different name server.
I have a server with limited memory. NSD was running out of memory. So I
tried Knot. Knot consumed similar memory (from a non-developer view, not
capably of analyzing virtual vs resistent vs ... memory).
So, I switched the Bind and was happy because my monitoring now tells me
"50% memory used" - and I am not worried about memory anymore. Sometimes
I have time to investigate and analyze problems, but most of the time I
do not have time for debugging and just need a solution for my problem,
hence I still use Bind on this server.
Regards
Klaus