Conference 7.286::fddi

Here's some estimated data.  For example, to compare Ultrix IP
against VMS QIO would certainly make the VMS driver look very
inefficient.  Just be careful to compare apples and apples.  The
DEMFA driver about the same as any other VMS LAN driver in terms 
of efficiency.

                               (ESTIMATED)           (ESTIMATED)
                            ---- 64-bytes ----    --- 4500-bytes ---
                            Pk/Sec   MBits/Sec    Pk/Sec   MBits/Sec
                            ------   ---------    ------   ---------
Raw hardware performance    190000      100        2800       100
(a system can attain this
with sufficient computes)

VCI/FFI VMS driver performance
 6000-410                     8000        4        2800       100
 6000-510                    14000        7        2800       100
 6000-610                    29000       15        2800       100
 9000                        32000       16        2800       100

QIO performance
 6000-410                     1500        0.8      1200        43
 6000-510                     2600        1.3      2050        74
 6000-610                     5500        2.8      2800       100
 9000                         6000        3.0      2800       100

Note that VCI/FFI is the interface used by LAT and Clusters.  DECnet uses
a slightly slower interface called ALTSTART.  The slowest by far of all is
QIO which is the interface used by user applications.  The additional 
overhead is what makes the interface robust enough to be used safely
by user applications which are generally not performance oriented - if
they are, they would use ALTSTART or VCI/FFI which is harder to interface
to but much faster.

These numbers are estimates based on data gathered so far.  More complete
data will be forthcoming soon including CPU utilization (obviously, any
numbers less than 100 mbits/sec the CPU is 100% utilized).

For detailed hardware performance data, contact Ram TALLIS::Kalkunte for
a performance analysis.

Thank you for your quick and detailed answer. Now I understand what the 
performance will be and depend on packet size.
Can I ask a little bit more?
What do you mean by "to compare Ultrix IP against VMS QIO would certainly 
make VMS driver look very inefficient. Just be careful to compare apples to 
apples."?
I understand that "Ultrix IP is same thing as ALTSTART in DECnet. VMS QIO is 
on top of DECnet. You can compare Ultrix IP performance with ALTSTART, VMS 
QIO performance with Ultrix TCP socket interface or Ultrix UDP socket 
interface. Don't compare Ultrix IP with VMS QIO. That are apples and oranges."
Right?
Will UCX TCP/IP QIO interface be same thing as VMS QIO interface?

That sounds about right.  I don't know much about the Ultrix side of things
so the point was just to ensure that you were comparing the same things.

For details on the UCX IP QIO interface, contact Alex LASSIE::Conta.

- Dick

Thanks a lot for your help.

Inmaru Tetu

    I'm in a performance situation with a DECnet solution versus a third
    party TCP/IP solution.
    
    Rather than use COPY and a stopwatch, I told him to use DTS/DTR for a
    DECnet test (should help a bit).  What can I do with segment size,
    buffer sizes, number buffers, or other parameters to help level this
    playing field?  The configuration is 2 6310's with DEMFA's simply
    connected via a concentrator.  
    
    Your input is appreciated greatly.
    
    John Lynch
    VMS partner, not a network partner ;-)

    Make sure the4 account/proces you run the DTS and DTR program from,. as
    well as the NETACP process are endowed with MEGA quota,.. expcially
    for BYTLM, BIOLM, DIOLM, ENQLM, ASTLM, ...
    
    I have an account that has everythnig maxed out for these kinds of
    tests.
    
    							/Bill
    

    I ran DTS/DTR and got good enough results (1.6MB/sec).  The customer
    has to provide 1 MB/sec file transfer however ... a bit of a tougher
    problem.  The DTR/DTS at least showed us that the systems are capable
    of >1MB/sec.
    
    Using VMS COPY, I'm only getting 340KBytes/sec.  Using Multinet, the
    customer sees about 800 KBytes.  Should I:
    		
    	1. expect better perf. from copy?
    	2. try our UCX ftp to get the 1MB/sec
    	3. punt
    
    thanks,
    - John -

    There is a little (a lot?) of extra overhead in the file copy
    compared to the DTS/DTR. That is primarily due to the encoding/decoding
    of the DAP protocol for the copy.
    
    I wouldn't expcet it to be that much of an isue,.. are you running
    the file copys from that same "max quota'ed" account? Is the rest
    of the system (NETACP's quptas etc.) the same for DTS/DTR and the
    file copy?
    
    							/Bill
    

    this has been taken off-line.
    
    Thanks,

Is there a performance report which formalizes the data presented in .1?

I found what I needed in the Digital Technical Journal, Summer 1991 edition.

However - I'd still like to get an online pointer to the articles, if there
is one.

    I believe all the DTJ articles can now be found in VTX DTJ. Is that
    good enough?
    
    Debbie