Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 20 Dec 2008 10:52:18 +0900
From:      Pyun YongHyeon <pyunyh@gmail.com>
To:        Craig Butler <craig001@lerwick.hopto.org>
Cc:        freebsd-sparc64@freebsd.org
Subject:   Re: Tier 2 Support
Message-ID:  <20081220015218.GD79449@cdnetworks.co.kr>
In-Reply-To: <1229690593.81803.8.camel@main.lerwick.hopto.org>
References:  <089f01c96137$1b7ea530$527bef90$@com> <1229624249.72316.12.camel@buffy.york.ac.uk> <1229690593.81803.8.camel@main.lerwick.hopto.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Dec 19, 2008 at 12:43:13PM +0000, Craig Butler wrote:
 > 
 > > Are you talking about the Blade 100's (the desktop-style machines) or
 > > the B100 Blades (Sun's early attempt at Blade servers)?  If you're
 > > talking the former, I can confirm that FreeBSD runs very well on them,
 > > however I don't believe the on-board network card is supported on the
 > > latter.
 > 
 > aye, 
 > 
 > If its the b100s that slot into a b1600 chassis FreeBSD doesn't
 > currently work with the cassini nics that are in the chassis.... yet 
 > 
 > I am in the process of working with openbsd developers to try and port

It seems that 8KB sized Rx buffer should be used to support one of
nice feature(e.g. header spiliting) of Cassini. Cassini does not
allow multi DMA segment for a Rx buffer so it looks hard to get 8KB
sized buffer on 4KB PAGE_SIZE architectures without resorting to
large contiguous buffer preallocated in driver. Due to the nature
of header spliting it would require a lot of buffers(e.g. 2K or 4K
buffer entries in Rx descriptor ring) to get maximum performance
from Cassini. Maintaing local buffers with contiguous memory would
be very inefficient and it could be miserably fail due to lack of
contiguous memory. Did you solve the issue?
If we implement reasonable sized static local buffers in driver as
we did to support jumbo frames it would make Cassini work but that
approach will require a lot of copy operation as well as disabling
header spliting and will make Cassini poor performance controller.

 > theirs over.  Unfortunately work is getting in the way, our product is
 > being released to the public shortly and its all hands to the pumps.
 > 
 > Craig B
-- 
Regards,
Pyun YongHyeon



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20081220015218.GD79449>