[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: GNU/kFreeBSD "bits", publicity, status report



On 07/10/14 23:53, Christoph Egger wrote:
>>   - suspend/resume seems to work now
> 
> FWIW not with wpa_supplicant alive and never twice in a row here (though
> the first time seems to be reliable otherwise). But suspend/resume
> certainly is something noone really expects to have reliable on every
> hardware I guess

I wonder if pm-suspend could be ported, or we could otherwise quickly
write something similar to do certain things before suspend and after
resume.

>>   - Petr rewrote our threads implementation in GNU libc
> 
> Wasn't that in wheezy already? at least all the ruby hangs are gone in
> wheezy (and gone completely since squeeze EOL on the buildds)

Petr first shared NPTL threads here, which was after wheezy release:
https://lists.debian.org/debian-bsd/2013/07/msg00060.html

Some causes of thread-related hangs with LinuxThreads got fixed I think.
 Other cases just got worked around such as by not running problematic
tests (python), or setting a time limit on running the whole testsuite
(ruby).

>>   - virtio drivers for KVM (see next point)
> 
> Peter Palfrader asked about virtio stress-testing (especially on AMD
> hosts) -- I hope to get to that. Did your BigV system use virtio?

Is this for DSA?  BigV is operated by Bytemark in York;  they also
donated the HP blade servers that Debian has as KVM hosts there.  (It's
why I picked BigV for that experiment ;)

Yes I was using virtio.  The BigV compute nodes seemed to be AMD Opteron
although IIRC use something like NBD rather than physically attached
disks.  I probably wasn't doing heavy disk I/O because it was doing CPU
or memory-bound compile and linking mostly.  Also ZFS, with lz4
compression;  and sync=disabled on the whole zpool ("evil tuning").

>>   - we chose to keep sysvinit as default, but are in a position to
>>     easily use alternate init systems;  OpenRC looks promising for the
>>     future (perhaps usable already?);  I guess file-rc still works
> 
> Hmm we should really follow up on openrc and see how much already
> works. Would be great to have the potential jessie+1 default working in
> jessie. I'll look into that.

It might simply work, but otherwise it may be still useful for jails.
The very first time I got it working on kfreebsd, it was able to start
up a jail with it, very easily.

Regards,
-- 
Steven Chamberlain
steven@pyro.eu.org


Reply to: