Yes, 1746 has always struck me as an odd number, but the macro definitions which calculate it were puzzling enough that I didn't want to assume.  As I recall, there was commentary about fidgeting with the number which I also didn't think I understood at the time.  But this was over three years ago, so I could certainly be misremembering.

Just as a reminder for any Tops-20 user: your ERROR.SYS file will fill up with class 5 events (buffer not large enough) if you don't condition NTMAC to ignore this.  At NMXFIL+1, you need to change three lines, viz:

NMXFIL: $SFIL
          $FIL(2,4)     ;[T152] ;Event classes 2 through 4 known by monitor
          $NOFIL(5,5)   ;[T152] ;HECnet overflows us with packet too long
          $FIL(6,6)     ;[T152] ;Event class 6 known by monitor
          $NOFIL(^D96,^D96)     ;LCG specific events - turned off
IFN FTDEBUG <
          $FIL(^D480,^D480)     ;Accept event class 480 while debugging
>
        $EFIL

On 7/15/24 9:04 PM, Paul Koning wrote:
I think that 1476 is a TOPS-20 private thing, the data link buffer sized used by the Ethernet driver.  The NCP parameters control what routing advertises (buffer size) and what NSP asks for as connection segment size (segment buffer size).

	paul

On Jul 15, 2024, at 8:49 PM, Johnny Billquist <bqt@softjar.se> wrote:

Maybe someone can explain to me why it looks like the buffer size reported by NCP is 576 when it sounds like you are both setting it to 1476?

 Johnny

On 2024-07-15 22:48, Thomas DeBellis wrote:
These are the relevant lines from my SYSTEM:7-1-CONFIG.CMD:
   DECNET BUFFER-SIZE 1476
   DECNET MAXIMUM-ADDRESS 1023
   NODE VENTI2 2.522
   DECNET ROUTER-LEVEL-1
   ETHERNET 0 DECNET
The maximum buffer size of 1476 is what I determined from a review of the relevant code in the monitor, which I commented about, some months back (but see below).
This is the response from my local excutor (VENTI2::)
   OPR>ncp shOW exECUTOR chaRACTERISTICS
   Request # 280; Show Executor Node Characteristics Completed
   Executor Node = 2.522 (VENTI2)
      Identification = Venti Due Test System
      Management Version = 4.0.0
      CPU = DECSYSTEM1020
      Software Identification = Tops-20 7.1 PANDA II
      Loop Count = 1
      Loop Length = 127
      Loop With = Mixed
      Incoming Timer = 30
      Outgoing Timer = 60
      NSP Version = 4.0.0
      Maximum Links = 65535
      Delay Factor = 48
      Delay Weight = 10
      Inactivity Timer = 120
      Retransmit Factor = 10
      Routing Version = 2.0.0
      Type = Routing IV
      Routing Timer = 600
      Broadcast Routing Timer = 40
      Maximum Address = 1023
      Maximum Circuits = 20
      Maximum Cost = 100
      Maximum Hops = 16
      Maximum Visits = 20
      Maximum Broadcast Nonrouters = 64
      Maximum Broadcast Routers = 32
      Maximum Buffers = 80
      Buffer Size = 576
      Segment Buffer Size = 576
The buffer size of 576 is what I remember being for non-Ethernet packets going over an IMP (which I'm sure is irrelevant because this was IP only).  1500 is the maximum for non-jumbo sized Ethernet frames, the 1476 length being 24 bytes less.  I plan to investigate that when I finally get into the guts of why ERROR.SYS fills up with class 5 errors from NTMAN for packets being too long.
------------------------------------------------------------------------
On 7/14/24 5:04 AM, Johnny Billquist wrote:
Anders...

On 2024-07-14 10:20, Anders Andersson wrote:
"merlyn drforbin" <kropotkin@gmx.com> wrote:

Any way to set endpoint buffer size on panda decnet phase IV
Could  some of this  (in SYSTEM:7-CONFIG.CMD) be what you are looking for?
This is from RARITY (not PANDA, but DEC vanilla distribution,kind of), I'm
not sure I have  the correct values myself, been  experimenting.
         ; DECnet parameters:
;default is DECNET BUFFER-SIZE 576
decnet buffer-size 1476
;default is DECNET MAXIMUM-BUFFERS 80
;decnet maximum-buffers 40
Hmm. That do not seem to be the values currently used:

.ncp tell rarity sho exec cha

Node characteristics as of 14-JUL-24 10:58:22

Executor node = 1.802 (RARITY)

   Identification = Rarity of Maneframe Six
   Management version = 4.0.0
   Software identification = DECnet-20 Version 4.0
   Loop count = 1, Loop length = 127
   Loop with = Mixed
   Incoming timer = 30, Outgoing timer = 60
   NSP version = 4.0.0
   Maximum links = 65535, Delay factor = 48
   Delay weight = 10, Inactivity timer = 120
   Retransmit factor = 10, Routing version = 2.0.0
   Type = Routing IV, Routing timer = 600
   Broadcast routing timer = 40
   Maximum address = 1023, Maximum circuits = 20
   Maximum cost = 100, Maximum hops = 16
   Maximum visits = 20, Maximum broadcast endnodes = 64
   Maximum broadcast routers = 32, Maximum buffers = 80
   Buffer size = 576, Segment buffer size = 576

.


However, also to clarify some things. The segment buffer size should probably stay at 576. This is basically the size of packets used at the DECnet level, and I have some vague recollection that nodes should use the same size, or DECnet might fail. DECnet is not that clever about different sizes there.
The buffer size on the other hand can be larger (modulo VMS bug that I mentioned a few days ago), but that would only be used by link layer protocols, like the routing information packets that are sent, for example.

But if you are experimenting, please let me know if you figure something out. Positive or negative. Always good with more information.

  Johnny

_______________________________________________
HECnet mailing list -- hecnet@lists.dfupdate.se
To unsubscribe send an email to hecnet-leave@lists.dfupdate.se
-- 
Johnny Billquist                  || "I'm on a bus
                                 ||  on a psychedelic trip
email: bqt@softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol
_______________________________________________
HECnet mailing list -- hecnet@lists.dfupdate.se
To unsubscribe send an email to hecnet-leave@lists.dfupdate.se
_______________________________________________
HECnet mailing list -- hecnet@lists.dfupdate.se
To unsubscribe send an email to hecnet-leave@lists.dfupdate.se