Netseer: stream event telemetry notes for programmable data plane

NetSeer： Stream event telemetry for programmable data plane

significance

In a cloud provider with a large and complex network , To configure 、 Software and hardware errors and failures are inevitable . therefore , Applications often suffer from constant network performance anomalies , For example, the bandwidth decreases 、 Long tail delay 、 bandwidth / Delay jitter, etc

As applications become more performance sensitive , Abnormal network performance is becoming more and more important for cloud providers .

Challenge

It is very challenging to quickly mitigate abnormal network performance , Because it is in the coverage of network monitoring 、 There are very high standards for speed and accuracy .

In most network performance exceptions , The bottleneck is the time it takes to find out the cause of abnormal network performance . To quickly locate the cause , Cloud providers must be able to immediately discover all performance critical events that occur on application traffic , For example, packet loss 、 Congestion, etc , These are the direct triggers of abnormal network performance . Besides , The granularity of these events should be at the traffic level rather than at the traditional interface or device level , Because precise information about where and how the victim's application traffic is disrupted can significantly speed up cause determination . therefore , The cloud provider must monitor the data plane for streaming events , And make event detection as responsive as possible .

Unfortunately , The existing network monitoring methods are far from meeting the above requirements for troubleshooting abnormal network performance , Because they are not designed for this purpose .

programme

Recent advances in the commodity programmable data plane provide the basis for implementing monitoring logic directly at the starting point of the switch data plane itself . This method can achieve fine granularity and low cost .

This paper introduces a network monitor based on stream event telemetry NetSeer, It can be continuous 、 Monitor all individual streams simultaneously , And comprehensively detect flow events , Including packet discarding 、 congestion 、 Path changes and package pauses , The cost is moderate .NetSeer The core idea of is to detect convection events , Completely compress the event data into the programmable data plane , Then upload to the management plane , This can greatly reduce the monitoring overhead , While maintaining the fine granularity of the data .

There are the following innovations

• NetSeer Use the programmability of both sides of the link to cooperate to find the drop events between switches , And recover the stream information of discarded or damaged packets

• NetSeer Take full advantage of data plane programmability , Accurately identify packets that encounter events , Aggregate packets that experience the same flow level events , Lossless compression of convection events , Minimize the amount of event data generated

• NetSeer Through a novel ring packet driven event collection data plane design , Batch process small event messages into large data packets , To reduce the transmission and processing overhead of the switch

• NetSeer Adopt distributed mode to telemetry stream events , It can expand linearly with the network scale

• NetSeer Ensure zero false positives in event generation , And use with ASIC Switch for unloading CPU Discover and eliminate false positives with less overhead .

Open questions

summary

In this paper , We proposed NetSeer, This is a network monitor in the data plane , It can accurately capture stream level data plane events through the high-level programmable data plane , For the convenience of npa Troubleshooting . We use novel designs , Fully explore the flexibility and performance of the programmable data plane , prove NetSeer Almost full stream event coverage can be achieved , It has good scalability for network scale , And high accuracy of stream event data . Through case studies and experiments , We revealed NetSeer Powerful ability to accelerate and mitigate various network performance anomalies .