Conference 7.286::fddi

    I don't claim to be any sort of expert, but I just attended Satish
    Rege's presentation on the DEMFA (The Ultimate Performance FDDI
    Adapter). According to one of his slides, DEMFA performance improves
    from around 45Mbps (22 byte frames) to 100Mbps (>90 byte frames) on a
    VAX 6000. I don't know how the 5900 performs compared to a 6000 though.
    Suffice it to say the adapter is not going to be the bottleneck. You
    might be able to get more info from Satish.
    
    Debbie
     

	
	Thanks for your thoughts and the pointer......I will contact
	Satish and get more details....

	So long

Note the DEMFA does not have a common bus with the DS5900.  

      
      Not a direct answer to your question but......
    
      Our current ETHERNET systems can perform better than this !
    
      The DECworld "Bricks" demo can show a DECstation 5000 achieving 9.5 Mbps 
      over Ethernet (with only a little "cheating") and 30 Mbps on FDDI and 
      this is limited by the cpu.
    
      Stuff that up the competition !
    
      Rog

    The previous couple of replies point out that the competitor is BSing...
    
    However, Satish's numbers are a result of tests run with TOXIF. TOXIF
    is not a full blown operating system like VMS. TOXIF is a program
    that runs on a VAX and consumes the entire machine. Nothing else (VMS,
    U*X, applications etc) can run while TOXIF runs. TOXIF has one
    function, to stress the DEMFA. All of the available CPU power of the
    machine is consumed by TOXIF and used to run the DEMFA. Therefore, the
    numbers Satish presents show what the adapter can do when it is dirven
    with virtually infinte CPU horsepower. A Vax, running VMS, with 2000
    users will not have "infinte" CPU power to drive the DEMFA, and so the
    number could be lower.
    
    What I'm trying to say is that the performance is much better than the
    competitor says, but may not be as good as Satish says for a VAX/VMS
    system running under any significant load.
    
    							/Bill
    

Fair enough.  But there was also a report (using field test software, as 
I recall) of 60 Mb/s file access speed to a MSCP-served disk across FDDI.
In other words, through VAXcluster stuff and up to an application was running
at that rate.  Not bad...

	paul

    Oh yeah,.. the numbers are better than the competitor quoted,.. as I
    said. 60 Mbps aint bad,. as you say,. but its less than the 98 Mbps
    or whtaever the number is that Satish will quote via TOXIF.
    
    							/Bill
    

	I have not read this notes file for a while and didn't realize I was
	quoted.  In the next reply I will post some application level
	performance for DEMFA with UCX as mesuared by Alex Conta. You can
	see that VUP for VUP, DEMFA at system level is better than twice
	anything else that is out there. Best known competitor's system is
	Silicon Graphics with 80 VUP CPU getting 48 Mbits/sec. But first
	reply to 412.7

>From 412.7

>    Oh yeah,.. the numbers are better than the competitor quoted,.. as I
>   said. 60 Mbps aint bad,. as you say,. but its less than the 98 Mbps
>  or whtaever the number is that Satish will quote via TOXIF.

	Bill,
		I have never claimed 98 Mbits/sec. to be the application
	layer performance. It is adapter hardware and host bus hardware
	level performance. Debbie,  having just heard my seminar once, was
	able to understand it and clearly states in her note 412.1 
        "Suffice it to say the adapter is not going to be the bottleneck." 

	I hope you will make an attempt to understand how and why performance
	is quoted at different levels and how to interpret it. By
	writing a long winded answer in 412.5 and an obivious answer in
	412.7, you have unnecessarily confused the heck out of the readers.

	/satish

From:	LADDIE::CONTA        "Alex Conta - DEC TCP/IP for VMS" 27-JAN-1992 21:18:32.22
To:	@FDDI_PERF
CC:	CONTA
Subj:	I:VAX 6620 TCP/IP FDDI on the COMNET floor

News from the COMNET floor....

After fixing a minor problem with the VMS TCP/IP, the DEC FDDI booth is
up and running full throttle, ready for the tomorrow's opening of the show. 
While I was over the phone, Kelly Caudill has fired two and then later three
DECStations 5000/200 all transmitting to a VAX 6620, running the UCX V2.0 EFT
and the "bricks" demos were showing the following: 

UDP/IP (32K datagrams)

2 DECstations xmit at around 30-32Mbit/sec -> VMS rcv 62.5 Mbit/sec
3 DECstations xmit at around 30-32Mbit/sec -> VMS rcv 78-80 Mbit/sec

TCP/IP (32K user data, 64K windows)

2 DECstations xmit at around 18-19Mbit/sec -> VMS rcv 37-38 Mbit/sec
3 DECstations xmit at around 17-18Mbit/sec -> VMS rcv 53-54 Mbit/sec

At this point the VAX 6600 VMS TCP/IP numbers look faster than the Silicon
Graphics that I have seen at INTEROP 91, which was the fastest at that show. 

Alex

    Satish,
    
    	The numbers are pretty good no matter how you look at it.
    
    	You are somewhat overzealous in your statements to the effect
    that the adapter is not the bottleneck. Truly the adapter can receive
    and transmit at remendous rates. BUT, simply transmitting and
    receiveing are only the rudimentary tasks. Even doing these rudimentary
    tasks require some aomount of CPU. The CPU can become the bottleneck
    if the device requires too much work for a simple transmit. Even if the
    amount of CPU to do one transmit is small,. when you start doing
    thousands of transmissions per second, you start to consume more
    CPU,.. perhaps more than is available. 
    
    	So yes,.. the adaopter is not a bottleneck,. but everyone should
    remember that adapters consume CPU if they are soing any useful work.
    Too much CPU/transmit,.. or too many transmits,. and the CPU can
    become overlaoded.
    
    	Whats the bottleneck? Perhaps if the device neede less CPU per
    transmit, the CPU would not be the bottleneck.
    
    							/Bill