Details
-
Spike
-
Not Assigned
-
None
-
None
-
Obs Mgt & Controls
-
-
-
2
-
7.5
-
Sprint 5
-
-
-
16.3
Description
We consider a short timeboxed Spike of a Sprint to look at the measurement possibilities and As Is performance. Further work could be covered later (A cloned Feature SP-712 created to capture all the initial descriptions)
Serious performance issues have been detected when a realistic number of CSP sub-arrays and other TANGO Servers and Devices are instantiated. The cause has not been identified, it may be inherent to pyTango implementation, caused by sub-optimal container configuration, blocking due to extensive use of forwarded attributes, use of synchronous (as opposed of asynchronous commands) commands.
In this Spike the team can look at baselining the current performance limits under a very similar deployment architecture replicating as closely as possible the number and hierarchy (of Pods, Containers, Tango Devices) as expected in CSP.LMC and document the results. If possible extend and scale till the performance degrades seriously or devices crash. This should be run on the Engage cluster.
Some of the description below may not occur this Spike but could be part of the follow up feature:
These are some of the options that will be explored:
- Where appropriate replace synchronous with asynchronous commands (in particular when command completion depends on large number of other devices or involves parsing of large JSON objects).
- Experiment using CSP.LMC and Mid.CBF images that instantiate large number of TANGO Servers/Devices.
- Experiment with different container configurations.