[prev in list] [next in list] [prev in thread] [next in thread] 

List:       unbound-users
Subject:    libunbound performance
From:       Liang Zhu via Unbound-users <unbound-users () nlnetlabs ! nl>
Date:       2019-04-02 21:55:26
Message-ID: CAC3z6OiEXT0-_+_cKm3CfZyASQ4prOiBQ5c7XgYEryLgtBG0Sw () mail ! gmail ! com
[Download RAW message or body]

Hello,

Does anyone have any numbers for libunbound benchmark? For example,
how many queries (not cached) per second can you achieve with
libunbound?

I have been playing with libunbound under Windows. It works well but
the query rate I can achieve is a couple of hundred queries (not
cached) per second. If I send at a faster rate, query latency will be
very large (3 seconds or more). I tried the instructions on the
"howto-optimise" page, but the result does not improve much.

In my test, cpu and ram are not overloaded and usage is low. All the
query names are unique and I have a wildcard record in my test zone.
The libunbound binary is downloaded from unbound website. The network
connection at my test machine is fast and normally it take less than
40ms to resolve the test record with cold cache.

My test program runs on 64bit-Windows and it uses:
- one ub_ctx
- a few threads keep calling ub_resolve_async()
- a few threads wait on ub_wait() for callback

One thing I have not tried yet is to build libunbound from source with
libevent to unblock "outgoing-range" options.

Any suggestions on improving the performance while using libunbound?
Is there any performance numbers you can share?

Thanks,
-Liang
[prev in list] [next in list] [prev in thread] [next in thread]