nmav at gnutls.org
Wed Jun 4 11:38:55 CEST 2014
On Mon, Jun 2, 2014 at 6:18 PM, Kurt Roeckx <kurt at roeckx.be> wrote:
>> It would really help to have some cachegrind/perf results or a way to
>> reproduce that. I'm working on some optimizations in libtasn1, and that
>> could be part of it.
> This are all public CRLs that I downloaded. You can see a list
> of CRLs at https://isc.sans.edu/crls.html. You can pick the size
> that you want there.
I've optimized some things in libtasn1 but the times were so bad that
are insignificant. 50% of the time is spent on decoding (I guess that
is expected as a large tree, is a large tree), and the other 50% on
reading the individual values. My change gave a speedup on the
decoding part by caching some values and drops it to 30% of the total
However, to get further improvements an API change is required, i.e.,
an iterator based function to read values for large sequences or sets
that are read in a serialized way.
More information about the Gnutls-devel