Advance in datacenter networking in the past decade has driven a sea change in the way datacenters are organized and managed. We are exploring various datacenter networking issues from the systems perspective. Currently, we are interested in low-latency, RDMA-based network systems.
Many datacenter applications such as machine learning and streaming systems do not need the complete set of data to perform their computation. Current approximate applications in datacenters run on a reliable network layer like TCP and either sample data before sending or drop data after receiving to improve performance. These approaches are network oblivious and transmit (and retransmit) more data than necessary, affecting both application runtime and network bandwidth usage.
We propose to run approximate applications on a lossy network and to allow packet loss in a controlled manner. We designed a new network protocol called Approximate Transmission Protocol, or ATP, for datacenter approximate applications. ATP opportunistically exploits available network bandwidth as much as possible, while performing a loss-based rate control algorithm to avoid bandwidth waste and retransmission. It also ensures bandwidth fair sharing across flows and improves accurate applications’ performance by leaving more switch buffer space to accurate flows.
Recently, there is an increasing interest in building datacenter applications with RDMA because of its low-latency, high-throughput, and low-CPU-utilization benefits. However, RDMAis not readily suitable for datacenter applications. It lacks a flexible, high-level abstraction; its performance does not scale; and it does not provide resource sharing or flexible protection. Because of these issues, it is difficult to build RDMA-based applications and to exploit RDMA’s performance benefits.
To solve these issues, we built LITE, a Local Indirection TiEr for RDMA in the Linux kernel that virtualizes native RDMA into a flexible, high-level, easy-to-use abstraction and allows applications to safely share resources. Despite the widely-held belief that kernel bypassing is essential to RDMA’s low-latency performance, we show that using a kernel-level indirection can achieve both flexibility and lowlatency, scalable performance at the same time.
MemAlbum: an Object-Based Remote Software Transactional Memory System
Shin-Yeh Tsai, Yiying Zhang
the 2018 Workshop on Warehouse-scale Memory Systems (WAMS '18) (co-located with ASPLOS '18)
LITE Kernel RDMA Support for Datacenter Applications
Shin-Yeh Tsai, Yiying Zhang
To appear in the Proceedings of the 26th ACM Symposium on Operating Systems Principles (SOSP '17)
Rockies: A Network System for Future Data Center Racks
Shin-Yeh Tsai, Linzhe Li, Yiying Zhang
WIP and Poster at the 14th USENIX Conference on File and Storage Technologies (FAST '16)