4. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 4
Stateful
Singelton
Component
(SSC)
SSC- a single threaded stateful component responsible to handle
certain functionalities of Ericsson RBS architecture
Challenges
5. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 5
› The SSC of Ericsson RBS have significantly long
session duration
› In addition to a kind of interaction where the SSC will
act as a server, SSC will also act as a client
› SSC should be responsive during switchover
challenges
6. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 6
RQ1 - Is it possible to have controlled switchover
functionality for SSC in RBS to preserve component
functionality and state, and thus increase system
availability?
RQ2 - What changes are needed compared to the
current way of implementing SSC to better support
switchover?
RQ3 - What extra functionality needs to be added in SSC
to handle switchover?
Research questions
7. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 7
› Qualitative research
– design patterns to ease switchover process
– session and state replication techniques, and object serialization
› Design research
Approaches
9. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 9
› SUSE Linux Enterprise Server
› Enea LINX Interprocess Communication Platform
› Plab Messaging Library
› Signal Distributer Library
Approaches---Setting
10. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 10
Replication States
System States
Results --- Prototype design
states
12. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 12
› Serialization : the process of translating an
object into a format that can be stored/restored
› Four technologies investigated:
– Data Object Library
– Protocol Buffers
– Boost
– Cereal (the choosen one)
Results
Object serialization
13. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 13
› C++11 header only serialization library
› Has different representation
❖compact binary encodings
❖XML
❖JSON
› 100% performance advantage over boost and
Protocol Buffers
› Easier to comprehend and extend
Cereal
17. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 17
Results
Prototype design strategies
Name serving
18. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 18
Results
Prototype design strategies
Service request handling
19. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 19
Results
Prototype design strategies
Service response handling
› LINX only guarantees FIFO
ordered messages
› Message Order
Consistency with Sequence
Numbers
R1R1
R2R2
R3R3
G G
2
1
1
3
3
2
Preparation phase
20. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 20
Results
Prototype design strategies
Service response handling
21. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 21
System state
Results --- Prototype design
states
22. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 22
› Correctness test via predetermining the expected
behavior during and after switchover process,
› Stress test of #Groups and #payload on
switchover process duration
› Stress test of #missed indication messages on
switchover process duration for synch purpose
› Stress test of #Groups on serialization and
deserialization duration
› Efficiency test on memory utilization during
serialization
Evaluation of the prototype
26. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 26
Time Effectiveness of serialization and
deserialization
27. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 27
Efficiency test on memory utilization during
serialization
28. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 28
➔Two additional Processes
➔Two additional Interface Units
➔~45 bytes memory per group during Se/Deserialization
➔Messages
◆ Switchover process management
◆ Transport serialized data
◆ Subscription and unsubscription for resource states
◆ Synchronization of missing indication messages
Discussion
(RQ2 & RQ3)
Added functionalities
29. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 29
~1K LOC
~100 LOC
› ~5 LOC on Class definitions for Cereal
› Linking of Cereal Library
Discussion
(RQ2 & RQ3)
Added functionalities
30. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 30
➔Switchover process duration grows linearly
with number of groups
◆ May change when tested with real product
environment
➔Cereal is efficient when compositional
object relations are serialized together
Discussion
performance
32. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 32
Future work-- suggested design
pattern
33. HA-Thesis Reference Component Model | Ericsson Internal | 2016-02-05 | Page 33
conclusion
› Potential implications towards strategies for high availability
Relations between no of groups vs size of each group vs switchover-duration
Each group has assigned 2 baseband and 2 resources
Each resource will send Ind msg. every five seconds
We notice that :
payload doesn’t have significant effect on the switchover duration
But the number of groups has significant effect on switchover duration
Relationship between synchronization of missing msgs vs switchover-duration
Relationship between numbers of objects vs time consumption
Up to 300 groups are created and serialized
We observe that serialization time is negligible and almost the same as the group number increased
On the contrary, the time consumed during deserialization is linearly increased.
Up to 30 groups are created
Serializing each group separately has a huge effect on memory utilization