☀️

Hello World.

Papers of Interest in ATC 22

ATC 22 has 18 tracks, and 64 papers are accepted.

  1. Distributed Systems 1/2
  2. Machine Learning 1/2/3
  3. Disaggregated Systems (what are the differences between disaggregated system and distributed system?)
  4. Networking 1/2
  5. Deployed Systems 1/2 (don’ t know what is )

Paper titles in ATC are more interesting than those in SIGCOMM or INFOCOM.

Interested Papers

Investigating Managed Language Runtime Performance: Why JavaScript and Python are 8x and 29x slower than C++, yet Java and Go can be Faster?

This paper compares the performance of Java, Go, JavaScript and Python with C++.

We found that V8/Node.js and CPython exhibit excessive overheads, executing applications 8.01x and 29.50x slower on average than their C++ counterparts.
In contrast, OpenJDK and Go applications are performance competitive to C++, running only 1.43x and 1.30x slower.

Features of managed language:

  • Easier readability and usability
  • Dynamic type checking
  • Memory management with garbage collection
  • Dynamic memory safety checks
Programming Language Applications
Java Hadoop
Java Spark
Go Kubernetes (a distributed key-value store)
Go M3 (a distributed time series database and query engine built by Uber)
Go Biscuit
Python Openstack
Python Paypal
Python Instagram
Python Dropbox

Experimental Results

Firebolt: Finding Bugs in Programmable Data Plane Generators

为了解决P4语言编程困难的问题,很多研究者研究制作P4代码生成器。这篇文章提出了一个工具为代码生成器debug。

We apply Firebolt to three popular open-source DP generators, i.e., Marple, Sonata, and Poise.

I’m interested in these DP generators.

Campo: Cost-Aware Performance Optimization for Mixed-Precision Neural Network Training

这篇论文考虑了,神经网络训练时,梯度量化带来的精度损失问题。

在网内聚合中,由于可编程交换机的限制,工作节点需要将梯度转换成整型才能让可编程交换机进行聚合。

Co-opting Linux Processes for High-Performance Network Simulation

针对现有网络仿真器的缺点进行优化,提出一个新的网络仿真器。对标NS-3 以及 Mininet。和我们做的工作相关度为0。

, — Sep 16, 2022