RE: Early results of Using the Channel Access Nameserver..

[Jeff Hill <johill@lanl.gov>]

Matthias,

I don't think that this is the very simple name server solution
that we discussed in Vancouver (i.e. the one in
"src/cas/example/directoryService").

My favorite solution is currently a multi-tier name resolution
system where one or more distributed caches are interrogated first,
and then the normal CA mechanisms are used only when the name isn't
found in the cache(s). Clients would update the cache(s) when they
discover a new or changed PV location via broadcasted (or normally
addressed) search requests. Configuration options would determine
whether several hosts share one cache or if there is one cache
for each host (or both). Servers would perhaps also update the
cache system when they boot. Of course none of this is implemented
at this time.

No doubt that the first step should be to create a plug in interface
in the client library so that sites can easily substitute the
name resolution solution that fits best with their situation.
This plug in interface should allow for information to flow both directions.
That is, the client should be able to interrogate the PV address
service, and the client should be able to update the PV address
service when it receives information on the PV's location directly
from the PV's server via search requests.

Jeff

On Friday, September 05, 1997 12:26 AM, Matthias Clausen DESY -MKS-2/KRYK- 
[SMTP:CLAUS@vxhera.desy.de] wrote:
> Is this the result of the discussion in Vancouver?
> Namely: is this nameserver based on the port. ca-server and Jeffs add-ons to
> CA?
> If yes:
> - you probably use the EPICS_CA_ADDR_LIST to find the nameserver!?
> - do you also use another EPICS_CA_SERVER_PORT?
> - how do you configure the nameserver?
>   - does it broadcast itself (at least once)
>   - is it based on static files?
> - ... and does a ca_search fall through to broadcast if the nameserver is not
> found?
>
> Matthias