1. NoSQL Stores for Coreless Mobile Networks
Frans Ojala+, Ashwin Rao*, Hannu Flinck‡, and Sasu Tarkoma*
+Avarko OY *University of Helsinki ‡Nokia Bell Labs
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

2. BACKGROUND: LTE Overview
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

3. Large number of signals exchanged during procedures
UE STATE INFORMATION
+ Add variable
x Remove variable
U Update variable
No changes
Large number of signals exchanged during procedures
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

4. Trends FOR NETWORK optimization
Move functionality to the network Edge
Example: Extensions of Mobile Edge Computing (MEC)
Move functions to the network Core
Example: Cloud Radio Access Network (CRAN)
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

5. Example CORELESS MOBILE NETWORK
Using a data store for keeping the UE state?
This is in line with the idea of having storage functions such as UDSF for UE state in 5G (TS )
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

6. NOSQL STORES FOR UE STATE
At least one identifier (e.g., IMSI) does not change value during the lifetime of a subscriber
This identifier can be the key
The other variables can be the value
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

7. HigH-Availability and CONSISTENCY (HAC)
Key Observation:
UE is present in only one geographical location at one point of time
Can we relax the availability and consistency criteria of UE state in locations far away from the current geographical location?
The data will be consistent and highly available in geographical locations close to the UE
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

8. HIGH AVAILABILITY AND CONSISTENCY Zones (AN EXAMPLE)
Data is highly available and consistent in HAC zones
Data is eventually consistent in Global zone
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

9. COMPARISON OF NOSQL STORES
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

10. BENCHMARKING APACHE GEODE FOR LTE WORKLOADS (GEODE TOPOLOGY)
The Locator orchestrates the cluster
The Serving Nodes store the data and respond to queries.
Emulate LTE workloads using Yahoo! Cloud Serving Benchmark (YCSB)
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

11. Experiment Setup Devices Used
5 Dell C6320 Servers
Two CPUs operating at 2.5GHz with a total of 48 cores
256 GB RAM
10 Gbps Ethernet
Ubuntu 16.04, Apache Geode M2, Java 1.8
Disconnected from the Internet to avoid cross traffic.
Workload: YCSB to emulate traffic from 1 million UEs, each making 3 million operations
Metrics: Throughput (op/s) and Latency (sec)
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

12. TEST Scenarios Partitioned: Redundant copies with primary
Writes to primary copy and synchronously propagated to secondary copes. Reads to any copy.
Replicated: Redundant copies without primary
Writes performed at all nodes simultaneously, client receives ack from each copy. Reads to any copy
HAC
Delta and no-delta updates in each scenario
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

13. RESULTS (PARTIONED) No delta updates Delta updates
NoSQL stores opens possibility for x improvements
The non-delta update mechanism performs better than delta updates
Apache Geode’s current mechanism incurs high processing overhead because of a small UE object
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

14. OTHER Results (presented in PAPER)
latency
throughput
throughput
latency
PARTIONED
No delta updates
Delta updates
REPLICATED
HAC
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

15. KEY OBSERVATIONS
The partitioned region demonstrated good scaling properties with the addition of new servers
Replicated regions throughput decreases when new serving nodes and servers were added
The non-delta update mechanism performs constantly better than delta updates
Apache Geode’s current mechanism incurs high processing overhead because of a small UE object
Our implementation of HAC performed poorly because we were limited by synchronization of YCSB threads
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

16. CONCLUDING REMARKS
A coreless mobile network opens avenues for meeting some of the requirements of future mobile networks
We explore how the eventual consistency property of NoSQL stores can be leveraged for creating High Availability and Consistency zones
We discuss the shortcomings and benefits of using a NoSQL store for managing the state of UEs in a coreless mobile network
Opens possibility for x improvement
We plan to explore approaches for implementing HAC-zoning for meeting requirements from various verticals
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao

17. http://ncar.cs.helsinki.fi ashwin.rao@helsinki.fi
THANK YOU
NoSQL Stores for Coreless Mobile Networks
Ashwin Rao