Re: /dev/crypto versus libcrypto

[Matthew Mondor <mm_lists%pulsar-zone.net@localhost>]

On Mon, 23 Jul 2012 08:52:39 -0400
Greg Troxel <gdt%ir.bbn.com@localhost> wrote:

> (For those reading further down, I find that mounting the 2T disk w/o
> rump and w/o wapbl and not using either coprocessor results in a stable
> system).

Nice page.

I remember there being performance and stability issues with rump and
puffs (especially on netbsd-5, not necessarily on Soekris).  It's
likely that there are leaks as well.  It's also possible that the large
size you observed could be due to caching.

I've not seen particular issues with WAPBL on netbsd-5, though.  I have
some servers in production using it with netbsd-5 and
kern.maxvnodes = 65536.  It's quite possible that it doesn't behave as
well with some PATA/SATA controllers or via USB, however...  Those
production systems are using piixide(4).  My current "desktop" also
uses WAPBL on netbsd-6/amd64 with ahcisata(4) "Intel 6 Series AHCI"
fine it seems.

It'd be nice to know what's wrong with WAPBL if your system is more
stable without it.  But the remaining errors mostly appear to to be
USB/umass0 related on your page (and it's not surprising that WAPBL
would change the device access behaviour of course).  My experience
with USB storage devices on NetBSD is limited to occasional use, I have
no permanently connected USB storage.
-- 
Matt