[Ukfreebsd] [FreeBSD-Announce] FreeBSD Status Report January-March, 2010 (fwd)

[Robert Watson]

On Fri, 23 Apr 2010, Sevan / Venture37 wrote:

>>> Are there any plans for reviving FreeBSD/sun4v??
>> 
>> I'm not aware of active work in this area, although the sparc64 folks are 
>> clearly getting stuff done on related architectures. I'm not sure what it 
>> would take to motivate further improvement on sun4v: the basic port 
>> appeared to work moderately well but there were stability issues relating 
>> to VM, and I think it wasn't yet able to run as a logical domain (perhaps 
>> have the wrong terminology). So there's a very strong starting point for 
>> anyone who wanted to pick this up as a project, but no active hands 
>> currently I think.
>
> It would be really good to see it up & running properly, what's the maximum 
> number of cores the freebsd kernel has been tested on? The OpenBSD guys 
> really ran with the support for this platform, the support for logical 
> domains was really cool, being able to mix operating systems on a single box 
> is very impressive, though one thing I don't get is why is a sun4v a 
> seperate platform on FreeBSD & a part of sparc64 in OpenBSD??

While I have great respect for the OpenBSD folk, I don't think a non-parallel 
kernel can be considered to "really run" on sun4v, stable or otherwise. 
There's little point in having 64 hardware threads if your kernel can only run 
on one at a time, especially with network-centric workloads.  My on-going work 
on network stack scalability for Juniper is targeted at 16-core amd64 servers 
and 32-thread MIPS network appliance parts (I'm using RMI/NetLogicMicro XLS 
and XLR).  You can read a bit about it here:

   http://www.freebsd.org/news/status/report-2010-01-2010-03.html#TCP/UDP-connection-groups

The goal is to effectively utilize 16 cores or 32 threads in concurrent TCP 
processing with at least 10gbps, and ideally 20gbps.

Robert