Optimizing remote procedure calls in datacenters using hardware/software co-design
Abstract/Contents
- Abstract
- In this dissertation, we present the nanoPU, a new NIC-CPU co-design that provides ultra low and predictable remote procedure call (RPC) response time and thus accelerates datacenter applications. The nanoPU achieves its goal by providing a fast path between the network and applications. This fast path has the following three characteristics: (1) it moves key resource scheduling decisions from software to hardware (reliable network transport & congestion control, RPC load balancing across cores, thread scheduling) allowing them to operate much more efficiently, (2) it provides a path directly between the network and applications which bypasses the cache and memory hierarchy placing arriving messages directly into the CPU register file, and (3) it supports a unique thread scheduling feature to bound the tail response time experienced by certain high-priority applications. We built an FPGA prototype of the nanoPU fast path by modifying an open-source RISC-V CPU, and evaluated its performance using cycle-accurate simulations on AWS FPGAs. The wire-to-wire time for nanoPU to receive an incoming message and initiate transmission of a response (response time) is just 69ns, an order of magnitude quicker than the best-of-breed, low latency, commercial NICs. Our hardware implementation of the NDP transport protocol adds less than 10ns to the wire-to-wire response time. We demonstrate that the hardware thread scheduler is able to lower (and potentially bound) RPC tail response time by about 5x while enabling the system to sustain 20% higher load, relative to traditional thread scheduling techniques. Furthermore, the nanoPU's core selection policy in hardware is able to efficiently distribute RPCs across cores eliminating hot spots and reducing tail response time. We implement and evaluate a suite of applications on the nanoPU, including MICA, Raft, and set algebra for document retrieval.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2021; ©2021 |
| Publication date | 2021; 2021 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Ibanez, Stephen Gabriel |
|---|---|
| Degree supervisor | McKeown, Nick |
| Thesis advisor | McKeown, Nick |
| Thesis advisor | Brebner, Gordon |
| Thesis advisor | Ousterhout, John K |
| Degree committee member | Brebner, Gordon |
| Degree committee member | Ousterhout, John K |
| Associated with | Stanford University, Department of Electrical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Stephen Ibanez. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2021. |
| Location | https://purl.stanford.edu/zb117jy9174 |
Access conditions
- Copyright
- © 2021 by Stephen Gabriel Ibanez
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...