PostMan: Rapidly Mitigating Bursty Traffic by Offloading Packet Processing
Talkat USENIX ATC '19 · on Friday, 2019-07-12· at 09:55· USENIX ATC Track II: Grand Ballroom VII–IX·
summary
System offloads CPU-heavy per-packet work to on-demand middleboxes called helpers that assemble small packets into large ones, replace TCP/IP headers with a 3-field PostMan header, and forward batched frames to the hot server. Implementation uses DPDK and mTCP; helpers stay stateless and support fast failover and connection migration. Evaluation shows Memcached mitigation in ~500 ms versus tens of seconds for data migration, 3.3x throughput for Memcached (2.8x for Paxos), and ~0.4 ms reconnection for 1,000 clients.
AI-written summary · about
chapters (15)
- 0:00Start
- 0:27Peak shopping season is going global
- 1:00Bursty traffic is a headache!
- 1:27Traditional remedy: migrating hot data for load balancing
- 2:49How to assemble small packets?
- 3:52Is batching in helpers efficient?
- 5:44Stateless failover mechanism
- 7:40How to enable helpers?
- 8:22Load balancing across helpers
- 9:29How to disable helpers?
- 10:26Complementary to data migration
- 11:15Evaluation Setup
- 12:14Performance gain of Post Man
- 12:44Fault tolerance
- 13:50Conclusion
related talks
16mQueue-Mem: Energy-Efficient Hardware Storage for Advanced Network Function Acceleration
21mAccelTCP: Accelerating Network Applications with Stateful TCP Offloading
16mPacket Order Matters! Improving Application Performance by Deliberately Delaying Packets
44mLinux Networking with MANA: RX Path Optimization and Netsh...
14mPoseidon: Efficient, Robust, and Practical Datacenter CC via Deployable INT
17mRearchitecting the TCP Stack for I/O-Offloaded Content Delivery
02h 53mPreview Sessions
28mMultipath TCP hands-on on Fedora