[bind10-dev] gprof results

[JINMEI Tatuya / 神明達哉]

At Tue, 8 Jun 2010 11:19:57 -0500 (CDT),
"Jeremy C. Reed" <jreed at isc.org> wrote:

> http://bind10.isc.org/~jreed/profiling/
> 
> http://bind10.isc.org/~jreed/profiling/gprof.out-100000-sqlite
> is 100,000 queries from record from sqlite db.

Not surprisingly, the query processing logic is a major bottleneck,
followed by response rendering.

[16]    77.4    0.04   17.23   99997         AuthSrv::processMessage() [16]
                0.27   11.13   99997/99997       isc::datasrc::DataSrc::doQuery() [17]
                0.07    3.85   99997/99997       isc::dns::Message::toWire() [23]

From a bit closer look at callees, major contribution to the overhead
in query processing is string related operations.

The major bottleneck in message rendering seems to be name
compression, which is quite predictable, and is probably optimizable
even for generic data sources quite easily.

---
JINMEI, Tatuya