fixedsearch interval computation
This commit is contained in:
@@ -25,6 +25,27 @@ problem does not occur in IOCs because they are not yet based on the portable
|
||||
server library. This bug was first detected by Ken Evans in the production CA
|
||||
gateway at the APS.</p>
|
||||
|
||||
<h4>CA Client Library's Search Datagram Interval Exponential Backoff
|
||||
Flawed</h4>
|
||||
|
||||
<p>CA's search datagram interval exponential backoff should look like
|
||||
this:</p>
|
||||
|
||||
<p>30mS, 60mS, 120mS, 240mS,...</p>
|
||||
|
||||
<p>Or a similar behavior with higher initial delays resulting from a round
|
||||
trip time delay estimate greater than 30 mS. However, instead the backoff
|
||||
delays looked like this</p>
|
||||
|
||||
<p>30 mS, 30mS, 30mS, 30mS, 30mS, 30mS, 60mS, 120mS, 240mS ...</p>
|
||||
|
||||
<p>This also impacts what CA does when a channel disconnects or there is a
|
||||
beacon anomaly ( a new server event). The intent was to start the search
|
||||
delay for disconnected channels in these situations at 2 seconds but due to
|
||||
the above bug the delay was more like 64 mS. </p>
|
||||
|
||||
<p>This bug appears only in earlier versions of EPICS R3.14</p>
|
||||
|
||||
<center>
|
||||
<h2>Changes since 3.14.3</h2>
|
||||
</center>
|
||||
|
||||
Reference in New Issue
Block a user