Timertasks notes

TimerTasks： Time driven execution towards programmable data plane

TimerTasks: Towards Time-driven Execution in Programmable Dataplanes

significance

The current programmable data plane provides an event driven execution model based on matching operation paradigm . The action is triggered when the data packet defined according to the matching specification arrives . Any programmed logic is executed only when there are package Events , Otherwise, no action will be triggered .

In recent years , Programmable data planes are used to implement high-level applications that were never thought of before , Such as network cache 、 Consensus agreement, etc . The implementation of many such systems requires action without the arrival of data packets .

Time driven execution in the data plane can help protocols and applications transfer new functions to the data plane .

Challenge

The main challenge is to implement time driven execution on existing event driven hardware .

programme

TimerTasks It can be implemented as a new hardware extension of the programmable switch model .

Our main idea is to have a framework of periodic events , It uses periodic events to coordinate time-driven execution . Periodic events are generated by generating periodic packets ( Called the data plane scale ) To achieve .

The periodic event framework uses on-chip memory for all TimerTask Instance maintenance timer . For each data plane marker , The timer will be decremented , For expired timers , The corresponding operation will be performed .

TimerTask It's a new one P4 object , It can work with P4 The rest of the program is combined to form the required logic .TimerTask The compiler can first TimerTasks Add to the recurring events framework ( stay P4 To realize ), Then merge the framework with the rest of the program . Got P4 The program can be submitted to regular P4 compiler .

TimerTask It provides high-level abstraction and richer semantics, avoiding the use of hand-made matching action tables .

Open questions

When supporting a large number of TimerTask When an instance , This adds additional pipeline processing overhead . How to support a large number of TimerTask example .

TimerTask The compiler needs to become more complex to properly combine multiple TimerTask.

TimerTasks Should be supported locally on hardware .

summary

How time driven execution in the data plane helps protocols and applications move new functionality to the data plane , A new method called TimerTasks The abstraction of . It allows the use of extended P4 Language grammar to express the required time driven semantics . It is implemented on the existing event driven hardware TimerTask The original language , And three new applications show its practicability .