[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

159.0. "6000 and FDDI" by RTPSWS::DAVIS () Tue Oct 23 1990 12:52

    Customer claims to have heard from "friends in DEC" that the VAX 6000
    (specifically the 400 family) becomes CPU bound for n/w access with the
    FDDI adapter under development.
    
    Is there any info in this area or should I ignore the question and
    indicate that the product is on the way and much more will be known
    then?  Can I say that the FDDI adapter will NOT adversely affect
    performance or is this not known yet?
    
    Thanks.

T.R	Title	User	Date	Lines
159.1	System not limited by adapter	STAR::STOCKDALE	`Thu Oct 25 1990 16:43`	11
	The way to look at it is that the system is not limited by the adapter, whether that is a 6000 or 9000 or next generation system. The customer probably wants to do something with a system besides network access so he really only wants to allocate a portion of total computes for network related processing. In actual fact, a 9000-210 doing QIOs can only drive a DEMNA Ethernet adapter to 50% of capacity using minimum size packets. To get the FDDI adapter going to its capacity at minimum size packets will take a lot more VUPs never mind XMI bandwidth. -Dick
159.2	VAX 6000 FDDi adapter is not the culprit	TALLIS::REGE	`Fri Oct 26 1990 15:54`	48
	From 159.0 > Customer claims to have heard from "friends in DEC" that the VAX 6000 > (specifically the 400 family) becomes CPU bound for n/w access with the > FDDI adapter under development. If the perception here is that the FDDI adapter is responsible for making VAX 6000 CPU bound then that perception is wrong. A network adapter cannot adversely impact CPU performance unless it generates excessive interrupts(e. g. interrupts for every packet received) or it does not implement Data Link Layer functionality. DEXFA - the XMI to FDDI adapter for 6000 -minimizes host interrupts and implements Data Link Layer functionality same as DEXNA - an XMI to Ethernet adapter - which is the highest performing Ethernet adapter shipping today. Also, DEXNA and DEXFA leave very little Data Link Layer functionality to be implemented in the host. Then the question is what is the difference between the CPU power required to run Ethernet vs. CPU power required to run FDDI? FDDI has 30 times the packet carrying capacity of Ethernet and 10 times the bandwidth of Ethernet. This is because the smallest packet that can be transmitted on FDDI is approximately three times smaller than the smallest packet that can be transmitted on Ethernet. Also, as is well known, FDDI has 10 times more bandwidth than Ethernet. (Here I am neglecting collisions etc. for a first level of approximation). This thirty times packet carrying capacity and 10 times the bandwidth is an indication of CPU power required to run FDDI at its maximum speed. Following quote from Bill Hancock in Client/Server Computing (supplement to Digital review, October 15,1990) on pp. 18 summarizes the present state of art for network adapters. "In almost all situations network controller is the predominant bottleneck". We intended to break this tradition and make DEXFA not be the bottleneck even on a high performance local area network such as FDDI. Therefore DEXFA was designed to make sure that it will not be the bottleneck on any system that uses XMI as a bus. In fact DEXFA will not be a bottleneck even on the 9000 system, which uses XMI for its network connection, leave alone on a 6000 system. Does this help you? If you need any more explanation I will be happy to discuss this over the phone.