Detecting if LTE is the Bottleneck with BurstTracker Arjun Balasingam , Manu Bansal, Rakesh Misra, Kanthi Nagaraj, Rahul Tandra, Sachin Katti, Aaron Schulman
Diagnosing Poor Streaming Quality 360p 2
Diagnosing Poor Streaming Quality 360p 2
Diagnosing Poor Streaming Quality 360p My video quality is only 360p (1.5 Mbps). The cellular downlink must be slow. 🤕 2
Diagnosing Poor Streaming Quality 360p My video quality is only 360p (1.5 Mbps). The cellular downlink must be slow. 🤕 2
Diagnosing Poor Streaming Quality 360p But a speed test says the downlink is 10 Mbps. My video quality is only 360p (1.5 Mbps). What’s going on?! The cellular downlink must be slow. 😡 🤕 2
Diagnosing Poor Streaming Quality 360p But a speed test says the downlink is 10 Mbps. My video quality is only 360p (1.5 Mbps). What’s going on?! The cellular downlink must be slow. 😡 🤕 We can not conclusively determine if the cellular downlink is the bottleneck. 2
Network Topology of a Mobile App Where is the 🤕 bottleneck? Internet Core Cellular App Server Network Radio Access Network 3
Network Topology of a Mobile App Where is the 🤕 bottleneck? Internet Core Cellular App Server Network Radio Access slow server code? Network 3
Network Topology of a Mobile App Where is the 🤕 bottleneck? Internet Core Cellular App Server Network Radio Access slow server code? throttling at middlebox? Network 3
Network Topology of a Mobile App Where is the 🤕 bottleneck? Internet Core Cellular App Server Network Radio Access slow server code? throttling at middlebox? Network congestion? 3
Network Topology of a Mobile App Where is the 🤕 bottleneck? Internet slow app code? Core Cellular App Server Network Radio Access slow server code? throttling at middlebox? Network congestion? 3
Network Topology of a Mobile App Where is the 🤕 bottleneck? Internet slow app code? Core Cellular App Server Network Radio Access slow server code? throttling at middlebox? Network congestion? 3
A Developer’s Perspective of Network Bottlenecks Internet Core Cellular App Server Network Radio Access Network 4
A Developer’s Perspective of Network Bottlenecks y t i l i b i s i v o N Internet Core Cellular App Server Network Radio Access Network 4
The Ideal Metric Queues Clients Radio (wireless) Link Base Station 5
The Ideal Metric If empty, traffic bottlenecked on the way to the base station. Queues Clients Radio (wireless) Link Base Station 5
The Ideal Metric If empty, traffic bottlenecked on the way to the base station. Queues Clients Radio (wireless) Link Base Station If nonempty, bottleneck is at base station . 5
The Ideal Metric If empty, traffic bottlenecked on the way to the base station. Queues Clients Radio (wireless) Link Base Station If nonempty, bottleneck is at base station . Providers know the status of the queues; but no one else does. BurstTracker estimates this metric at the client. 5
The Ideal Metric Queues Clients Radio (wireless) Link Base Station Opportunity: base station scheduling is accessible at the client! Providers know the status of the queues; but no one else does. BurstTracker estimates this metric at the client. 5
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 6
Resource Allocation Reveals Queue Status Resources scheduled to purple user in this millisecond. Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 6
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 7
Resource Allocation Reveals Queue Status Base station gives all resources to one user! Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 7
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 8
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 8
Resource Allocation Reveals Queue Status Purple user not scheduled due to contention; its queue is still nonempty. Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 8
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 9
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 9
Resource Allocation Reveals Queue Status Purple user not given all resources; its queue has drained out. Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 9
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 10
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 A user only needs to know their resource allocation to infer their queue status. 11
Resource Allocation Reveals Queue Status Resource Block (RB) User Burst Time (msec) 1 Begin Likely 1 End 2 A user only needs to know their resource allocation to infer their queue status. 11
Does BurstTracker Generalize? Provider: Verizon 1 Mbps over 1 min Cumulative Probability 1.00 0.75 0.50 0.25 0.00 0 50 100 Resources allocated in each time slot 12
Does BurstTracker Generalize? Provider: Verizon 1 Mbps over 1 min Cumulative Probability 1.00 0.75 0.50 0.25 0.00 0 50 100 Resources allocated in each time slot 12
Does BurstTracker Generalize? Provider: Verizon 1 Mbps over 1 min Cumulative Probability 1.00 0.75 0.50 0.25 busy cell idle cell 0.00 0 50 100 Resources allocated in each time slot Slow transfer was aggregated into bursts that used most of the resources. 13
Does BurstTracker Generalize? Provider: Verizon 1 Mbps over 1 min Cumulative Probability 1.00 0.75 0.50 0.25 busy cell idle cell 0.00 0 50 100 Resources allocated in each time slot Slow transfer was aggregated into bursts that used most of the resources. 13
Does BurstTracker Generalize? Provider: Verizon 1 Mbps over 1 min Cumulative Probability 1.00 0.75 0.50 0.25 busy cell idle cell 0.00 0 50 100 Carrier Agg. SISO MIMO Resources allocated in each time slot Slow transfer was aggregated into bursts that used most of the resources. 13
Is BurstTracker Accurate? Experiment Setup BurstTracker Application • 100 runs of each workload Median Error (%) • Network conditions ~ 2–12.5 Mbps File Download 7.2 Partnered with Tier-1 provider to get ground-truth queue status measurements. Video Streaming 6.9 14
Is BurstTracker Accurate? Experiment Setup BurstTracker Application • 100 runs of each workload Median Error (%) • Network conditions ~ 2–12.5 Mbps File Download 7.2 Partnered with Tier-1 provider to get ground-truth queue status measurements. Video Streaming 6.9 BurstTracker achieves a median error of 7% for different classes of mobile apps. 14
Case Study: Video Streaming Throughput (Mbps) ● 6 ● ● ● speed test speed test speed test ● ● ● ● ● before ● 4 ● before after ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● speed test speed test speed test ● ● ● ● 2 average average ● average video streaming video streaming video streaming 0 0 50 100 150 200 time (s) We found that, surprisingly, the LTE downlink was not the bottleneck. 15
Case Study: Video Streaming Throughput (Mbps) ● 6 ● ● ● speed test speed test speed test ● ● ● ● ● before ● 4 ● before after ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● speed test speed test speed test ● ● ● ● 2 average average ● average video streaming video streaming video streaming 0 0 50 100 150 200 time (s) We found that, surprisingly, the LTE downlink was not the bottleneck. 15
1 Video Segment at the Client Resource Allocation Trace for a Single Video Segment Burst End Burst Start RBs/msec 1.00 Fraction of RBs 0.75 0.50 0.25 0.00 0 200 400 600 time (msec) 16
1 Video Segment at the Client Resource Allocation Trace for a Single Video Segment Burst End Burst Start RBs/msec 1.00 Fraction of RBs 0.75 0.50 0.25 0.00 0 200 400 600 time (msec) 16
1 Video Segment at the Client Resource Allocation Trace for a Single Video Segment Burst End Burst Start RBs/msec 1.00 Fraction of RBs 0.75 0.50 0.25 0.00 0 200 400 600 time (msec) BurstTracker indicates that it might be TCP Slow-Start. 16
Slow-Start Restart at the Middlebox HTTP (80) HTTPS (443) unreserved (7777) 100 Link Demand Ratio (%) 75 • Only HTTP and HTTPS traffic used middlebox. 50 • Nonstandard port (7777) bypasses middlebox. 25 0 AT&T T − Mobile Verizon 17
Recommend
More recommend
