1. FreeBSD Network Stack Performance Srinivas Krishnan University of North Carolina at Chapel Hill

2. Outline Introduction Unix network stack improvements Bottlenecks Memory Copies Interrupt Processing Zero Copy Implementation Receive Live Lock Solution

3. Introduction NIC IP Queue Packet Transport + Network Socket Queue Soft Interrupt Memory Copy Kernel Processing User Processing

4. Network Stack Reinvented Van Jacobson Net Channels Create a High Speed Channel from NIC to User space Push all processing to the user space Applying E2E “truly” Preserve cache coherency for multi- processor systems BETTER INTERRUPT PROCESSING

5. Network Stack Reinvented Ulrich Drepper’s Asynchronous Network I/O Asynchronous sockets True Zero Copy No Locking Event Channels BETTER MEMORY PROCESSING

6. Reduce Memory Copies Sending Side Copy from User Buffer to Kernel Buffer Copy from Kernel Buffer to Device Buffer Receive Side Copy from Device Buffer to Kernel Buffer Copy from Kernel Buffer to Socket Buffer

7. Zero Copy Send write Userspace Pages RAM Page Sized chunks External mbuf DMA into Driver Buffer NIC

8. Zero Copy Read NIC Packet Kernel Buffer DMA Kernel Space User Space User Buffer read(fd, buf, s)

9. Zero Copy Allocate an External Mbuf Pool NIC MTU has to be >= 4K Intel Pro1000 NIC with Jumbo Frames 3Com NIC turn on DMA Buffer and stitch the data together Added Overhead

10. Page Flipping Check Mbuf len Page Size ! 1 Page Use copyout Use vm_pgmovecovm_pgmoveco (……) Kernel Page User Page read(….) Atleast 1 Page

11. Preliminary Results 1500 bytes MTU (Iostat trace) for 10 mins

12. Processing Interrupts Main Processing Hard Interrupt from NIC to driver Soft Interrupt from IP Queue to processing Reduce user level and interrupt thread processing Problem: Receive Live Locks

13. Receive Live Lock Send large stream of UDP packets > receiver buffer capacity CPU spent processing network packets Goodput = 0

14. Implementation Design NIC IP Queue Packet Transport + Network Socket Queue Driver Queue Scheduler

15. Components All UDP packets are queued in driver queue Scheduler is triggered with the arrival of first UDP packet Checks the queue every n ms (currently 1-2ms) Schedules packet departure rate based on timestamps

16. Driver Queue Algorithm Set maximum rate and average rates Driver Queue maintains Average Queue Length (Weighted over time) Current Rate of transfer Time stamp of packets

17. Algorithm (cont) If current_rate > average rate Drop N packets such that current_rate == average_rate If current_rate > max rate (Spike) Drop all packets Reduce Time Wait in Queue If Current Queue Size < threshold Schedule packet exit such that rate == average_rate Appends an exit time to each packet

18. Pros and Cons Easy implementation requires no scheduling changes Reduces CPU utilization in worst case by ~25% Low Overhead Introduces added jitter

19. Experimental Setup Receive UDP Data Intel Pro1000 Nics Send UDP Data Intel Pro1000 Nics Iostat Trace Netstat trace Custom queue stats

20. Queue Stats At the Receiver Collect Average Queue Size CPU Utilization Packet Drops Total Number of packets processed

21. Receive Live Lock

22. Receive Live Lock (soln)

23. Future Work Feedback from Socket Queue and IP queue such that Weighted Average computed over all 3 queues Drop at driver before DMA Driver buffer not large enough to keep weighted queue size Feedback from Driver Queue Scheduler to driver to drop

24. Questions ?