Re: UDP search requests caget vs pvget

[Michael Davidsaver via Tech-talk <tech-talk at aps.anl.gov>]

On 3/16/21 8:17 AM, Stainer Tom via Tech-talk wrote:
> Dear,
> 
>  
> 
> After reading some documentation and presentations I came across the fact that Channel Access will send out UDP search packets at a 30 ms initial interval then doubling until some limit. However I could not find an equivalent algorithm for PVA.
> 
>  
> 
> I did some basic tests comparing caget with pvget for an invalid PV name to listen to the UDP search packets and indeed witnessed CA as described in the documentation (first dump), but for PVA it seems somewhat different, starting with a 200 ms interval and then roughly doubling until 60 seconds and then having an interval of 30/31 seconds thereafter (second dump).
> 
>  
> 
> Are my tests correct?
> 
>  
> 
> If so, can anyone please help me understand why PVA follows a different search pattern?

> Is it more efficient, or is it primarily done to not conflict with CA search requests?

I'm also not sure how to define "efficient" here.  From my experience,
both schemes tend to manifest as bursts of broadcasts.  In particular, imo.
the 30ms initial retry time of libca (this really is an implementation detail)
is too short.

It might be relevant that the libca scheme predates modern switched ethernet.
I've occasionally wondered if the shorter initial retry was motivated by
the some aspect of half-duplex ethernet.

fyi. for PVXS I'm using a scheme with a linear back off delay with the idea to
"leveling off" the period bursts of traffic while large numbers of channels
are disconnected.

Also fyi. https://github.com/mdavidsaver/cashark


...
> Apologies if this is already documented somewhere, but I have been unable to find information on this.

The details of the various client search algorithms (there are others) haven't
so far been documented in detail.