[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference 7.286::fddi

Title:	FDDI - The Next Generation

Moderator:	NETCAD::STEFANI

Created:	Thu Apr 27 1989
Last Modified:	Thu Jun 05 1997
Last Successful Update:	Fri Jun 06 1997
Number of topics:	2259
Total number of notes:	8590

1454.0. "802.2 LLC Performance from OSF/1" by AUSSIE::elstree.sna.dec.com::kimpton (Stark Raving Sane) Tue Sep 27 1994 05:05

Hi !

I'm curious as to whether anybody has every done any performance 
characteristics on the delivery of 802.2 LLC packets from a DEFTA 
interface (in a Flamingo/Pelican etc) using the example code provided 
with OSF/1 (ie /usr/examples/dli/dli_802.c).

Whilst I'm aware that in general the above configuration of 
hardware/Operating System can saturate (or as near to as possible) an 
FDDI ring - I'm wondering whether that was using code such as the 
above or whether it was from TCP/IP etc. etc.

Many Thanks

Andrew 8-)

T.R	Title	User	Personal Name	Date	Lines
1454.1		NETRIX::thomas	The Code Warrior	`Tue Sep 27 1994 08:53`	2
	I doubt it. The saturation testing was done from TCP or UDP. (OSG seems to ignore the raw datalinks).
1454.2		AUSSIE::elstree.sna.dec.com::kimpton	Stark Raving Sane	`Tue Sep 27 1994 19:24`	12
	Righto - that seems to bear out what I'm seeing (just about 66 MBit/s). I need to do some more checking into what I've done with the original 'example' but does that figure seem to be reasonable ? I thought it might be just a bit low for a fairly idle machine and a quiet network. Thanks Andrew 8-)
1454.3		koning.lkg.dec.com::koning	Paul Koning, B-16504	`Wed Sep 28 1994 10:56`	7
	That's surprising: given that TCP/UDP can saturate the link, you should be able to do likewise using raw datalink access. If you miss it by a large margin, as you indicated, then either (a) your application is quite inefficient, or (b) the raw interface is very inefficient. paul
1454.4		AUSSIE::elstree.sna.dec.com::kimpton	Stark Raving Sane	`Wed Sep 28 1994 23:31`	8
	The performance I'm getting at the moment is about adequate for my needs - and since other deadlines are looming I don't really have the time to investigate too much. Perhaps if things ease of in a couple of weeks - I'll re-investigate. Andrew 8-_
1454.5		NETRIX::thomas	The Code Warrior	`Thu Sep 29 1994 08:57`	2
	The one thing you should do is use the SO_RCVBUF setsockopt to increase your socket to ~64K.
1454.6	3000/500 defta performance info	SOLVIT::PAGE		`Tue Oct 25 1994 09:59`	10
	Running on a 3000/500 defta, I use about 10% of the CPU per 10 mega-bits FDDI on receive bandwidth. Therefore, I can just saturate the ring by taking 100% of the cpu. [btw, the OSF 3.0 driver has a curious bug, in that it will drop packets on receive if their length is approx 120 bytes, but if I use 4k packets, its ok.] ...Cal
1454.7		NETRIX::thomas	The Code Warrior	`Tue Oct 25 1994 10:56`	1
	What type of packet? 802.2 LLC? Ethernet? Padded Ethernet?
1454.8	here is bind data structure	SOLVIT::PAGE		`Tue Oct 25 1994 11:16`	14
	Looking at the structure I pass to bind, it is setup as follows: dli_family = AF_DLI dli_substructure = DLI_802 dli_802addr.ioctl = DLI_EXCLUSIVE dli_802addr.svc = TYPE1 dli_802addr.eh_802.ssap = SNAP_SAP dli_802addr.eh_802.dsap = SNAP_SAP dli_802addr.eh_802.ctl.U_fmt = UI_NPCMD; ... ...Cal
1454.9		NETRIX::thomas	The Code Warrior	`Tue Oct 25 1994 11:41`	2
	SNAP SAP eh? packets greater than 120 should work fine. I know that MOP uses them.