杂七杂八的，随手记录。

Linux Process States

yc 发布于 2023-02-01

Table of Contents

1 The Linux Process States
2 Checking Process State
- 2.1 Displaying Process State Using ps
- 2.2 Using the top Command
- 2.3 The /proc Pseudo File
3 Summary

本文为摘录(或转载)，侵删，原文为： https://www.baeldung.com/linux/process-states

1 The Linux Process States

Running or Runnable (R)
Uninterruptible Sleep (D)
Interruptable Sleep (S)
Stopped (T)
Zombie (Z)

状态机如下：

A Learned Query Rewrite System using Monte Carlo Tree Search

yc 发布于 2022-11-01

Table of Contents

1 INTRODUCTION
2 LearnedRewrite
- 2.1 树形结构
3 PRELIMINARIES
- 3.1 Query Rewrite Rules
- 3.2 Query Rewrite
4 TREE SEARCH FOR QUERY REWRITE
- 4.1 Overview of Policy Tree Search

本文为摘录(或转载)，侵删，原文为： attachments/pdf/7/p46-li.pdf

查询重写使用启发式算法来实现，有两个限制
- 规则的应用顺序严重影响查询性能，但
  - 可能的重写顺序随查询涉及到的算子指数增长
  - 受限于搜索空间大小限制，很难找到最佳的顺序
- 针对不同的查询，不同的重写规则的收益也不同
  - 当前的方法，只能应用于单个计划，而不能有效的估计查询重写的收益
提出了基于策略树树的查询重写框架

QueryFormer: A Tree Transformer Model for Query Plan Representation

yc 发布于 2022-11-01

Table of Contents

1 INTRODUCTION

本文为摘录(或转载)，侵删，原文为： attachments/pdf/0/p1658-zhao.pdf

QueryFormer:

learning-based query planer representation model
with tree-structured Transformer architecture

-integrate histograms from database into query plan encoding

1 INTRODUCTION

Physical Query Plan As DAG (Directed Acyclic Graph)
Figure 1: Example query and query plan
- node 表示操作
- edge 表示方向
- 子节点先执行，执行结果给父节点作为输入
Physical Query Plan 作为机器学习的输入，用以数据库优化

Adaptive Range Filters for Cold Data: Avoiding Trips to Siberia

yc 发布于 2022-10-01

Table of Contents

1 ABSTRACT
2 INTRODUCTION
3 APPLICATION EXAMPLE
- 3.1 Project Siberia

本文为摘录(或转载)，侵删，原文为： attachments/pdf/f/p1714-kossmann.pdf

对目前的架构来讲，和 粗糙索引 作用重叠。。

1 ABSTRACT

ARF: Adaptive Range Filter, 自适应范围过滤器
ARF is for Range queries, while
BloomFilter is for Point queries

2 INTRODUCTION

3 APPLICATION EXAMPLE

3.1 Project Siberia

Siberia 是 Hekaton 项目用于管理冷数据的一个项目
- Hekaton 则是 SQL Server 套件中的内存数据库

查询处理

ByteHTAP: ByteDance’s HTAP System with High Data Freshness and Strong Data Consistency

yc 发布于 2022-10-01

Table of Contents

1 INTRODUCTION
2 RELATED WORK
3 SYSTEM ARCHITECTURE
- 3.1 ByteNDB Overview
- 3.2 System Overview
- 3.3 Metadata Service
- 3.4 OLAP Engine
4 SHARED STORAGE WITH HIGH DATA FRESHNESS
- 4.1 Delta Store
- 4.2 Base Store
- 4.3 High Data Freshness
5 LSN-BASED STRONG DATA CONSISTENCY
6 OLAP QUERY PERFORMANCE OPTIMIZATION

本文为摘录(或转载)，侵删，原文为： ../pdf/5/p3411-chen.pdf

Chimp: Efficient Lossless Floating Point Compression for Time Series Databases

yc 发布于 2022-10-01

Table of Contents

1 ABSTRACT
2 INTRODUCTION
3 PRELIMINARIES
4 PROPERTIES OF REAL-WORLD TIME SERIES
- 4.1 Trailing Zeros
- 4.2 Leading Zeros
- 4.3 Revisiting Gorilla Compression
5 OVERVIEW
- 5.1 Our Chimp Algorithm
6 效果对比
- 6.1 Compression size result
- 6.2 Compression and decompression time

本文为摘录(或转载)，侵删，原文为： attachments/pdf/1/p3058-liakos.pdf

1 ABSTRACT

时序数据难以高效存储，导致存储代价高昂。

Coping with the TCP TIME-WAIT state on busy Linux servers

yc 发布于 2022-10-01

Table of Contents

1 TL;DR
2 About the TIME-WAIT state
- 2.1 TCP state diagram
- 2.2 Purpose
3 Problems
4 Other solutions
- 4.1 net.ipv4.tcp_tw_reuse
5 Another
6 总结

本文为摘录(或转载)，侵删，原文为： -https://draveness.me/whys-the-design-tcp-time-wait/ -https://vincent.bernat.ch/en/blog/2014-tcp-time-wait-state-linux#summary

1 TL;DR

不用启用 net.ipv4.tcp_tw_recycle ，该选项已在 Linux4.12 废弃
多数情况下， TIME-WAIT 状态的 socket 无害

2 About the TIME-WAIT state

2.1 TCP state diagram

http://tcpipguide.com/free/t_TCPOperationalOverviewandtheTCPFiniteStateMachineF-2.htm

Greenplum 部署最佳实践: 网络篇

yc 发布于 2022-10-01

Table of Contents

1 使用高速硬件，降低节点间的延迟
2 尽量降低网络中断的次数。
3 尽量提高每一个数据包的有效载荷

本文为摘录(或转载)，侵删，原文为： attachments/pdf/a/cn.greenplum.org-Greenplum部署最佳实践-网络篇.pdf

Performance of 10 G Ethernet Using Commodity Hardware

yc 发布于 2022-10-01

Table of Contents

1 INTRODUCTION
2 NETWORK PROCESSING IN THE LINUX KERNEL
- 2.1 Kernel Accounting
- 2.2 Packet Transmission 发送
- 2.3 Packet Reception 接收
- 2.4 Incorrect SoftIRQ Accounting
3 TESTBED AND KERNEL CONFIGURATION
4 UDP TRANSMISSION
5 TCP TRANSMISSION
6 TCP TRANSMISSION WITH ZERO-COPY
7 THE HARDWARE OFFLOAD FUNCTIONALITIES OF THE NETWORK ADAPTER
8 FIBRE CHANNEL TO 10 GIGABIT ETHERNET TESTS
9 CONCLUSIONS

本文为摘录(或转载)，侵删，原文为： attachments/pdf/3/performanceOf10GbE.pdf

SNARF: A Learning-Enhanced Range Filter

yc 发布于 2022-10-01

Table of Contents

1 ABSTRACT
2 INTRODUCTION
- 2.1 Range Filters
3 SNARF: A LEARNED FILTER
- 3.1 SNARF Description

本文为摘录(或转载)，侵删，原文为： attachments/pdf/a/p1632-vaidya.pdf

1 ABSTRACT

SNARF: Sparse Numerical Array-Based Range Filters 基于稀疏数组的范围过滤器
用于数值类型的范围过滤

2 INTRODUCTION

Filters
- are space efficient but appropriate