(window.webpackJsonp=window.webpackJsonp||[]).push([[392],{959:function(t,e,v){"use strict";v.r(e);var _=v(69),a=Object(_.a)({},(function(){var t=this,e=t.$createElement,v=t._self._c||e;return v("ContentSlotsDistributor",{attrs:{"slot-key":t.$parent.slotKey}},[v("h1",{attrs:{id:"时间序列数据库比较"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#时间序列数据库比较"}},[t._v("#")]),t._v(" 时间序列数据库比较")]),t._v(" "),v("h2",{attrs:{id:"overview"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#overview"}},[t._v("#")]),t._v(" Overview")]),t._v(" "),v("p",[v("img",{attrs:{src:"https://user-images.githubusercontent.com/33376433/119833923-182ffc00-bf32-11eb-8b3f-9f95d3729ad2.png",alt:"TSDB Comparison"}})]),t._v(" "),v("p",[t._v("*"),v("em",[t._v("表格外观启发自"),v("a",{attrs:{href:"https://towardsdatascience.com/how-to-select-time-series-db-123b0eb4ab82",target:"_blank",rel:"noopener noreferrer"}},[t._v("Andriy Zabavskyy: How to Select Time Series DB"),v("OutboundLink")],1)])]),t._v(" "),v("h2",{attrs:{id:"_1-已知的时间序列数据库"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_1-已知的时间序列数据库"}},[t._v("#")]),t._v(" 1. 已知的时间序列数据库")]),t._v(" "),v("p",[t._v("随着时间序列数据变得越来越重要，一些开源的时间序列数据库（Time Series Databases，or TSDB）诞生了。")]),t._v(" "),v("p",[t._v("但是，它们中很少有专门为物联网（IoT）或者工业物联网（Industrial IoT，缩写IIoT）场景开发的。")]),t._v(" "),v("p",[t._v("本文把IoTDB和下述三种类型的时间序列数据库进行了比较：")]),t._v(" "),v("ul",[v("li",[v("p",[t._v("InfluxDB - 原生时间序列数据库")]),t._v(" "),v("p",[t._v("InfluxDB是最流行的时间序列数据库之一。")]),t._v(" "),v("p",[t._v("接口：InfluxQL and HTTP API")])]),t._v(" "),v("li",[v("p",[t._v("OpenTSDB和KairosDB - 基于NoSQL的时间序列数据库")]),t._v(" "),v("p",[t._v("这两种数据库是相似的，但是OpenTSDB基于HBase而KairosDB基于Cassandra。")]),t._v(" "),v("p",[t._v("它们两个都提供RESTful风格的API。")]),t._v(" "),v("p",[t._v("接口：Restful API")])]),t._v(" "),v("li",[v("p",[t._v("TimescaleDB - 基于关系型数据库的时间序列数据库")]),t._v(" "),v("p",[t._v("接口：SQL")])])]),t._v(" "),v("p",[t._v("Prometheus和Druid也因为时间序列数据管理而闻名，但是Prometheus聚焦在数据采集、可视化和报警，Druid聚焦在OLAP负载的数据分析，因为本文省略了Prometheus和Druid。")]),t._v(" "),v("h2",{attrs:{id:"_2-比较"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_2-比较"}},[t._v("#")]),t._v(" 2. 比较")]),t._v(" "),v("p",[t._v("本文将从以下两个角度比较时间序列数据库：功能比较、性能比较。")]),t._v(" "),v("h3",{attrs:{id:"_2-1-功能比较"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_2-1-功能比较"}},[t._v("#")]),t._v(" 2.1 功能比较")]),t._v(" "),v("p",[t._v("以下两节分别是时间序列数据库的基础功能比较（2.1.1）和高级功能比较（2.1.2）。")]),t._v(" "),v("p",[t._v("表格中符号的含义：")]),t._v(" "),v("ul",[v("li",[v("code",[t._v("++")]),t._v("：强大支持")]),t._v(" "),v("li",[v("code",[t._v("+")]),t._v("：支持")]),t._v(" "),v("li",[v("code",[t._v("+-")]),t._v("：支持但欠佳")]),t._v(" "),v("li",[v("code",[t._v("-")]),t._v("：不支持")]),t._v(" "),v("li",[v("code",[t._v("?")]),t._v("：未知")])]),t._v(" "),v("h4",{attrs:{id:"_2-1-1-基础功能"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_2-1-1-基础功能"}},[t._v("#")]),t._v(" 2.1.1 基础功能")]),t._v(" "),v("table",[v("thead",[v("tr",[v("th",[t._v("TSDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("IoTDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("InfluxDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("OpenTSDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("KairosDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("TimescaleDB")])])]),t._v(" "),v("tbody",[v("tr",[v("td",[v("em",[t._v("OpenSource")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("+")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("+")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("SQL-like")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Schema")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("Tree-based, tag-based")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("tag-based")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("tag-based")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("tag-based")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("Relational")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Writing out-of-order data")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Schema-less")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Batch insertion")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Time range filter")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Order by time")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Value filter")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Downsampling")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Fill")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("LIMIT")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("SLIMIT")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("?")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Latest value")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("++")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])])])]),t._v(" "),v("p",[t._v("具体地：")]),t._v(" "),v("ul",[v("li",[v("p",[v("em",[t._v("OpenSource")]),t._v("：")]),t._v(" "),v("ul",[v("li",[t._v("IoTDB使用Apache License 2.0。")]),t._v(" "),v("li",[t._v("InfluxDB使用MIT license。但是，"),v("strong",[t._v("它的集群版本没有开源")]),t._v("。")]),t._v(" "),v("li",[t._v("OpenTSDB使用LGPL2.1，"),v("strong",[t._v("和Apache License不兼容")]),t._v("。")]),t._v(" "),v("li",[t._v("KairosDB使用Apache License 2.0。")]),t._v(" "),v("li",[t._v("TimescaleDB使用Timescale License，对企业来说不是免费的。")])])]),t._v(" "),v("li",[v("p",[v("em",[t._v("SQL-like")]),t._v("：")]),t._v(" "),v("ul",[v("li",[t._v("IoTDB和InfluxDB支持SQL-like语言。另外，IoTDB和Calcite的集成几乎完成（PR已经提交），这意味着IoTDB很快就能支持标准SQL。")]),t._v(" "),v("li",[t._v("OpenTSDB和KairosDB只支持Rest API。IoTDB也支持Rest API（PR已经提交）。")]),t._v(" "),v("li",[t._v("TimescaleDB使用的是和PostgreSQL一样的SQL。")])])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Schema")]),t._v("：")]),t._v(" "),v("ul",[v("li",[t._v("IoTDB：IoTDB提出了一种"),v("a",{attrs:{href:"http://iotdb.apache.org/zh/UserGuide/Master/Data-Concept/Data-Model-and-Terminology.html",target:"_blank",rel:"noopener noreferrer"}},[t._v("基于树的schema"),v("OutboundLink")],1),t._v("。这和其它时间序列数据库很不一样。这种schema有以下优点：\n"),v("ul",[v("li",[t._v("在许多工业场景里，设备管理是有层次的，而不是扁平的。因此我们认为基于树的schema比基于tag-value的schema更好。")]),t._v(" "),v("li",[t._v("在许多现实应用中，tag的名字是不变的。例如：风力发电机制造商总是用风机所在的国家、所属的风场以及在风场中的ID来标识一个风机，因此，一个4层高的树（“root.the-country-name.the-farm-name.the-id”）来表示就足矣。你不需要重复告诉IoTDB”树的第二层是国家名”、“树的第三层是风场名“等等这种信息。")]),t._v(" "),v("li",[t._v("这样的基于路径的时间序列ID定义还能够支持灵活的查询，例如：”root.*.a.b.*“，其中*是一个通配符。")])])]),t._v(" "),v("li",[t._v("InfluxDB, KairosDB, OpenTSDB：使用基于tag-value的schema。现在比较流行这种schema。")]),t._v(" "),v("li",[t._v("TimescaleDB使用关系表。")])])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Order by time")]),t._v("：")]),t._v(" "),v("p",[t._v("对于时间序列数据库来说，Order by time好像是一个琐碎的功能。但是当我们考虑另一个叫做”align by time“的功能时，事情就变得有趣起来。这就是为什么我们把OpenTSDB和KairosDB标记为”不支持“。事实上，所有时间序列数据库都支持单条时间序列的按时间戳排序。但是，OpenTSDB和KairosDB不支持多条时间序列的按时间戳排序。")]),t._v(" "),v("p",[t._v("下面考虑一个新的例子：这里有两条时间序列，一条是风场1中的风速，一条是风场1中的风机1产生的电能。如果我们想要研究风速和产生电能之间的关系，我们首先需要知道二者在相同时间戳下的值。也就是说，我们需要按照时间戳对齐这两条时间序列。因此，结果应该是：")]),t._v(" "),v("table",[v("thead",[v("tr",[v("th",[t._v("时间戳")]),t._v(" "),v("th",[t._v("风场1中的风速")]),t._v(" "),v("th",[t._v("风场1中的风机1产生的电能")])])]),t._v(" "),v("tbody",[v("tr",[v("td",[t._v("1")]),t._v(" "),v("td",[t._v("5.0")]),t._v(" "),v("td",[t._v("13.1")])]),t._v(" "),v("tr",[v("td",[t._v("2")]),t._v(" "),v("td",[t._v("6.0")]),t._v(" "),v("td",[t._v("13.3")])]),t._v(" "),v("tr",[v("td",[t._v("3")]),t._v(" "),v("td",[t._v("null")]),t._v(" "),v("td",[t._v("13.1")])])])]),t._v(" "),v("p",[t._v("或者：")]),t._v(" "),v("table",[v("thead",[v("tr",[v("th",[t._v("时间戳")]),t._v(" "),v("th",[t._v("时间序列名")]),t._v(" "),v("th",[t._v("值")])])]),t._v(" "),v("tbody",[v("tr",[v("td",[t._v("1")]),t._v(" "),v("td",[t._v("风场1中的风速")]),t._v(" "),v("td",[t._v("5.0")])]),t._v(" "),v("tr",[v("td",[t._v("1")]),t._v(" "),v("td",[t._v("风场1中的风机1产生的电能")]),t._v(" "),v("td",[t._v("13.1")])]),t._v(" "),v("tr",[v("td",[t._v("2")]),t._v(" "),v("td",[t._v("风场1中的风速")]),t._v(" "),v("td",[t._v("6.0")])]),t._v(" "),v("tr",[v("td",[t._v("2")]),t._v(" "),v("td",[t._v("风场1中的风机1产生的电能")]),t._v(" "),v("td",[t._v("13.3")])]),t._v(" "),v("tr",[v("td",[t._v("3")]),t._v(" "),v("td",[t._v("风场1中的风机1产生的电能")]),t._v(" "),v("td",[t._v("13.1")])])])]),t._v(" "),v("p",[t._v("虽然第二个表格没有按照时间戳对齐两条时间序列，但是只需要逐行扫描数据就可以很容易地在客户端实现这个功能。")]),t._v(" "),v("p",[t._v("IoTDB支持第一种表格格式（叫做align by time），InfluxDB支持第二种表格格式。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Downsampling")]),t._v("：")]),t._v(" "),v("p",[t._v("Downsampling（降采样）用于改变时间序列的粒度，例如：从10Hz到1Hz，或者每天1个点。")]),t._v(" "),v("p",[t._v("和其他数据库不同的是，IoTDB能够实时降采样数据，而其它时间序列数据库在磁盘上序列化降采样数据。")]),t._v(" "),v("p",[t._v("也就是说：")]),t._v(" "),v("ul",[v("li",[v("p",[t._v("IoTDB支持在任意时间对数据进行即席（ad-hoc）降采样。例如：一条SQL返回从2020-04-27 08:00:00开始的每5分钟采样1个点的降采样数据，另一条SQL返回从2020-04-27 08:00:01开始的每5分10秒采样1个点的降采样数据。")]),t._v(" "),v("p",[t._v("（InfluxDB也支持即席降采样，但是性能似乎并不好。）")])]),t._v(" "),v("li",[v("p",[t._v("IoTDB的降采样不占用磁盘。")])])])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Fill")]),t._v("：")]),t._v(" "),v("p",[t._v("有时候我们认为数据是按照某种固定的频率采集的，比如1Hz（即每秒1个点）。但是通常我们会丢失一些数据点，可能由于网络不稳定、机器繁忙、机器宕机等等。在这些场景下，填充这些数据空洞是重要的。数据科学家可以因此避免很多所谓的”dirty work“比如数据清洗。")]),t._v(" "),v("p",[t._v("InfluxDB和OpenTSDB只支持在group by语句里使用fill，而IoTDB能支持给定一个特定的时间戳的fill。此外，IoTDB还支持多种填充策略。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Slimit")]),t._v("：")]),t._v(" "),v("p",[t._v("Slimit是指返回指定数量的measurements（或者，InfluxDB中的fields）。")]),t._v(" "),v("p",[t._v("例如：一个风机有1000个测点（风速、电压等等），使用slimit和soffset可以只返回其中的一部分测点。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Latest value")]),t._v("：")]),t._v(" "),v("p",[t._v("最基础的时间序列应用之一是监视最新数据。因此，返回一条时间序列的最新点是非常重要的查询功能。")]),t._v(" "),v("p",[t._v("IoTDB和OpenTSDB使用一个特殊的SQL或API来支持这个功能，而InfluxDB使用聚合函数来支持。")]),t._v(" "),v("p",[t._v("IoTDB提供一个特殊的SQL的原因是IoTDB专门优化了查询。")])])]),t._v(" "),v("p",[v("strong",[t._v("结论：")])]),t._v(" "),v("p",[t._v("通过对基础功能的比较，我们可以发现：")]),t._v(" "),v("ul",[v("li",[t._v("OpenTSDB和KairosDB缺少一些重要的查询功能。")]),t._v(" "),v("li",[t._v("TimescaleDB不能被企业免费使用。")]),t._v(" "),v("li",[t._v("IoTDB和InfluxDB可以满足时间序列数据管理的大部分需求，同时它俩之间有一些不同之处。")])]),t._v(" "),v("h4",{attrs:{id:"_2-1-2-高级功能"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_2-1-2-高级功能"}},[t._v("#")]),t._v(" 2.1.2 高级功能")]),t._v(" "),v("table",[v("thead",[v("tr",[v("th",[t._v("TSDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("IoTDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("InfluxDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("OpenTSDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("KairosDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("TimescaleDB")])])]),t._v(" "),v("tbody",[v("tr",[v("td",[v("em",[t._v("Align by time")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Compression")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("MQTT support")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Run on Edge-side Device")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Multi-instance Sync")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("JDBC Driver")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("+")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Standard SQL")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Spark integration")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Hive integration")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Writing data to NFS (HDFS)")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Flink integration")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[v("strong",[t._v("++")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("-")])])])]),t._v(" "),v("p",[t._v("具体地：")]),t._v(" "),v("ul",[v("li",[v("p",[v("em",[t._v("Align by time")]),t._v("：上文已经介绍过，这里不再赘述。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Compression")]),t._v("：")]),t._v(" "),v("ul",[v("li",[t._v("IoTDB支持许多时间序列编码和压缩方法，比如RLE, 2DIFF, Gorilla等等，以及Snappy压缩。在IoTDB里，你可以根据数据分布选择你想要的编码方法。更多信息参考"),v("a",{attrs:{href:"http://iotdb.apache.org/UserGuide/Master/Data-Concept/Encoding.html",target:"_blank",rel:"noopener noreferrer"}},[t._v("这里"),v("OutboundLink")],1),t._v("。")]),t._v(" "),v("li",[t._v("InfluxDB也支持编码和压缩，但是你不能定义你想要的编码方法，编码只取决于数据类型。更多信息参考"),v("a",{attrs:{href:"https://docs.influxdata.com/influxdb/v1.7/concepts/storage_engine/",target:"_blank",rel:"noopener noreferrer"}},[t._v("这里"),v("OutboundLink")],1),t._v("。")]),t._v(" "),v("li",[t._v("OpenTSDB和KairosDB在后端使用HBase和Cassandra，并且没有针对时间序列的特殊编码。")])])]),t._v(" "),v("li",[v("p",[v("em",[t._v("MQTT protocol support")]),t._v("：")]),t._v(" "),v("p",[t._v("MQTT protocol是一个被工业用户广泛知晓的国际标准。只有IoTDB和InfluxDB支持用户使用MQTT客户端来写数据。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Running on Edge-side Device")]),t._v("：")]),t._v(" "),v("p",[t._v("现在，边缘计算变得越来越重要，边缘设备有越来越强大的计算资源。")]),t._v(" "),v("p",[t._v("在边缘侧部署时间序列数据库，对于管理边缘侧数据、服务于边缘计算来说，是有用的。")]),t._v(" "),v("p",[t._v("由于OpenTSDB和KairosDB依赖另外的数据库，它们的体系结构是臃肿的。特别是很难在边缘侧运行Hadoop。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Multi-instance Sync")]),t._v("：")]),t._v(" "),v("p",[t._v("现在假设我们在边缘侧有许多时间序列数据库实例，考虑如何把它们的数据上传到数据中心去形成一个数据湖。")]),t._v(" "),v("p",[t._v("一个解决方法是从这些实例读取数据，然后逐点写入到数据中心。")]),t._v(" "),v("p",[t._v("IoTDB提供了另一个选项：把数据文件增量上传到数据中心，然后数据中心可以支持在数据上的服务。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("JDBC driver")]),t._v("：")]),t._v(" "),v("p",[t._v("现在只有IoTDB支持了JDBC driver（虽然不是所有接口都实现），这使得IoTDB可以整合许多其它的基于JDBC driver的软件。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Standard SQL")]),t._v("：")]),t._v(" "),v("p",[t._v("正如之前提到的，IoTDB和Calcite的集成几乎完成（PR已经提交），这意味着IoTDB很快就能支持标准SQL。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Spark and Hive integration")]),t._v("：")]),t._v(" "),v("p",[t._v("让大数据分析软件访问数据库中的数据来完成复杂数据分析是非常重要的。")]),t._v(" "),v("p",[t._v("IoTDB支持Hive-connector和Spark-connector来完成更好的整合。")])]),t._v(" "),v("li",[v("p",[v("em",[t._v("Writing data to NFS (HDFS)")]),t._v("：")]),t._v(" "),v("p",[t._v("Sharing nothing的体系结构是好的，但是有时候你不得不增加新的服务器，即便你的CPU和内存都是空闲的而磁盘已经满了。")]),t._v(" "),v("p",[t._v("此外，如果我们能直接把数据文件存储到HDFS中，用Spark和其它软件来分析数据将会更加简单，不需要ETL。")]),t._v(" "),v("ul",[v("li",[t._v("IoTDB支持往本地或者HDFS写数据。IoTDB还允许用户扩展实现在其它NFS上存储数据。")]),t._v(" "),v("li",[t._v("InfluxDB和KairosDB只能往本地写数据。")]),t._v(" "),v("li",[t._v("OpenTSDB只能往HDFS写数据。")])])])]),t._v(" "),v("p",[v("strong",[t._v("结论：")])]),t._v(" "),v("p",[t._v("IoTDB拥有许多其它时间序列数据库不支持的强大功能。")]),t._v(" "),v("h3",{attrs:{id:"_2-2-性能比较"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_2-2-性能比较"}},[t._v("#")]),t._v(" 2.2 性能比较")]),t._v(" "),v("p",[t._v("如果你觉得：”如果我只需要基础功能的话，IoTDB好像和其它的时间序列数据库没有什么不同。“")]),t._v(" "),v("p",[t._v("这好像是有道理的。但是如果考虑性能的话，你也许会改变你的想法。")]),t._v(" "),v("h4",{attrs:{id:"_2-2-1-快速浏览"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_2-2-1-快速浏览"}},[t._v("#")]),t._v(" 2.2.1 快速浏览")]),t._v(" "),v("table",[v("thead",[v("tr",[v("th",[t._v("TSDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("IoTDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("InfluxDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("KairosDB")]),t._v(" "),v("th",{staticStyle:{"text-align":"center"}},[t._v("TimescaleDB")])])]),t._v(" "),v("tbody",[v("tr",[v("td",[v("em",[t._v("Scalable Writes")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("++")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Raw Data Query")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("++")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Aggregation Query")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("++")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Downsampling Query")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("++")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")])]),t._v(" "),v("tr",[v("td",[v("em",[t._v("Latest Query")])]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("++")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+-")]),t._v(" "),v("td",{staticStyle:{"text-align":"center"}},[t._v("+")])])])]),t._v(" "),v("h5",{attrs:{id:"写入性能"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#写入性能"}},[t._v("#")]),t._v(" 写入性能")]),t._v(" "),v("p",[t._v("我们从两个方面来测试写性能：batch size和client num。存储组的数量是10。有1000个设备，每个设备有100个传感器，也就是说一共有100K条时间序列。")]),t._v(" "),v("p",[t._v("测试使用的IoTDB版本是"),v("code",[t._v("v0.11.1")]),t._v("。")]),t._v(" "),v("h6",{attrs:{id:"改变batch-size"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#改变batch-size"}},[t._v("#")]),t._v(" 改变batch size")]),t._v(" "),v("p",[t._v("10个客户端并发地写数据。IoTDB使用batch insertion API，batch size从1ms到1min变化（每次调用write API写N个数据点）。")]),t._v(" "),v("p",[t._v("写入吞吐率（points/second）如下图所示：")]),t._v(" "),v("img",{attrs:{src:"https://user-images.githubusercontent.com/24886743/106254214-6cacbe80-6253-11eb-8532-d6a1829f8f66.png",alt:"Batch Size with Write Throughput (points/second)"}}),t._v(" "),v("center",[t._v("Figure 1. Batch Size with Write throughput (points/second) IoTDB v0.11.1")]),t._v(" "),v("p",[t._v("写入延迟（ms）如下图所示：")]),t._v(" "),v("p",[v("img",{attrs:{src:"https://user-images.githubusercontent.com/24886743/106251391-df1b9f80-624f-11eb-9f1f-66823839acba.png",alt:"Batch Size with Write Delay (ms)"}})]),t._v(" "),v("center",[t._v("Figure 2. Batch Size with Write Delay (ms) IoTDB v0.11.1")]),t._v(" "),v("h6",{attrs:{id:"改变client-num"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#改变client-num"}},[t._v("#")]),t._v(" 改变client num")]),t._v(" "),v("p",[t._v("client num从1到50变化。IoTDB使用batch insertion API，batch size是100（每次调用write API写100个数据点）。")]),t._v(" "),v("p",[t._v("写入吞吐率（points/second）如下图所示：")]),t._v(" "),v("p",[v("img",{attrs:{src:"https://user-images.githubusercontent.com/24886743/106251411-e5aa1700-624f-11eb-8ca8-00c0627b1e96.png",alt:"Client Num with Write Throughput (points/second) (ms)"}})]),t._v(" "),v("center",[t._v("Figure 3. Client Num with Write Throughput (points/second) IoTDB v0.11.1")]),t._v(" "),v("h5",{attrs:{id:"查询性能"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#查询性能"}},[t._v("#")]),t._v(" 查询性能")]),t._v(" "),v("p",[t._v("10个客户端并发地读数据。存储组的数量是10。有10个设备，每个设备有10个传感器，也就是说一共有100条时间序列。")]),t._v(" "),v("p",[t._v("数据类型是"),v("em",[t._v("double")]),t._v("，编码类型是"),v("em",[t._v("GORILLA")]),t._v("。")]),t._v(" "),v("p",[t._v("测试使用的IoTDB版本是"),v("code",[t._v("v0.11.1")]),t._v("。")]),t._v(" "),v("p",[t._v("测试结果如下图所示：")]),t._v(" "),v("p",[v("img",{attrs:{src:"https://user-images.githubusercontent.com/24886743/106251377-daef8200-624f-11eb-9678-b1d5440be2de.png",alt:"Raw data query 1 col"}})]),t._v(" "),v("center",[t._v("Figure 4. Raw data query 1 col time cost(ms) IoTDB v0.11.1")]),t._v(" "),v("p",[v("img",{attrs:{src:"https://user-images.githubusercontent.com/24886743/106251336-cf03c000-624f-11eb-8395-de5e349f47b5.png",alt:"Aggregation query"}})]),t._v(" "),v("center",[t._v("Figure 5. Aggregation query time cost(ms) IoTDB v0.11.1")]),t._v(" "),v("p",[v("img",{attrs:{src:"https://user-images.githubusercontent.com/24886743/106251353-d32fdd80-624f-11eb-80c1-fdb4197939fe.png",alt:"Downsampling query"}})]),t._v(" "),v("center",[t._v("Figure 6. Downsampling query time cost(ms) IoTDB v0.11.1")]),t._v(" "),v("p",[v("img",{attrs:{src:"https://user-images.githubusercontent.com/24886743/106251369-d7f49180-624f-11eb-9d19-fc7341582b90.png",alt:"Latest query"}})]),t._v(" "),v("center",[t._v("Figure 7. Latest query time cost(ms) IoTDB v0.11.1")]),t._v(" "),v("p",[t._v("可以看到，IoTDB的raw data query、aggregation query、downsampling query、latest query查询性能表现都超越了其它数据库。")]),t._v(" "),v("h4",{attrs:{id:"_2-2-2-更多细节"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_2-2-2-更多细节"}},[t._v("#")]),t._v(" 2.2.2 更多细节")]),t._v(" "),v("p",[t._v("我们提供了一个benchmark工具，叫做"),v("a",{attrs:{href:"https://github.com/thulab/iotdb-benchmark",target:"_blank",rel:"noopener noreferrer"}},[t._v("IoTDB-benchamrk"),v("OutboundLink")],1),t._v("（你可以用dev branch来编译它）。它支持IoTDB, InfluxDB, KairosDB, TimescaleDB, OpenTSDB。")]),t._v(" "),v("p",[t._v("我们有一篇文章关于用这个benchmark工具比较这些时间序列数据库："),v("a",{attrs:{href:"https://arxiv.org/abs/1901.08304",target:"_blank",rel:"noopener noreferrer"}},[t._v("Benchmarking Time Series Databases with IoTDB-Benchmark for IoT Scenarios"),v("OutboundLink")],1),t._v("。我们发表这个文章的时候，IoTDB才刚刚加入Apache incubator，所以我们在那篇文章里删去了IoTDB的性能测试。但是在比较之后，一些结果展示在这里：")]),t._v(" "),v("ul",[v("li",[t._v("对于InfluxDB，我们把cache-max-memory-size和max-series-perbase设置成unlimited（否则它很快就会超时）。")]),t._v(" "),v("li",[t._v("对于KairosDB，我们把Cassandra的read_repair_chance设置为0.1（但是这没有什么影响，因为我们只有一个结点）。")]),t._v(" "),v("li",[t._v("对于TimescaleDB，我们用PGTune工具来优化PostgreSQL。")])]),t._v(" "),v("p",[t._v("所有的时间序列数据库运行的机器配置是：Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz, (8 cores 16 threads), 32GB memory, 256G SSD and 10T HDD, OS: Ubuntu 16.04.7 LTS, 64bits.")]),t._v(" "),v("p",[t._v("所有的客户端运行的机器配置是：Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz，(6 cores 12 threads), 16GB memory, 256G SSD, OS: Ubuntu 16.04.7 LTS, 64bits.")]),t._v(" "),v("h2",{attrs:{id:"_3-结论"}},[v("a",{staticClass:"header-anchor",attrs:{href:"#_3-结论"}},[t._v("#")]),t._v(" 3. 结论")]),t._v(" "),v("p",[t._v("从以上所有实验中，我们可以看到IoTDB的性能大大优于其他数据库。")]),t._v(" "),v("p",[t._v("IoTDB具有最小的写入延迟。批处理大小越大，IoTDB的写入吞吐量就越高。这表明IoTDB最适合批处理数据写入方案。")]),t._v(" "),v("p",[t._v("在高并发方案中，IoTDB也可以保持吞吐量的稳定增长。 （每秒1200万个点可能已达到千兆网卡的限制）")]),t._v(" "),v("p",[t._v("在原始数据查询中，随着查询范围的扩大，IoTDB的优势开始显现。因为数据块的粒度更大，列式存储的优势体现出来，所以基于列的压缩和列迭代器都将加速查询。")]),t._v(" "),v("p",[t._v("在聚合查询中，我们使用文件层的统计信息并缓存统计信息。因此，多个查询仅需要执行内存计算（不需要遍历原始数据点，也不需要访问磁盘），因此聚合性能优势显而易见。")]),t._v(" "),v("p",[t._v("降采样查询场景更加有趣，因为时间分区越来越大，IoTDB的查询性能逐渐提高。它可能上升了两倍，这对应于2个粒度（3小时和4.5天）的预先计算的信息。因此，分别加快了1天和1周范围内的查询。其他数据库仅上升一次，表明它们只有一个粒度统计。")]),t._v(" "),v("p",[t._v("如果您正在为您的IIoT应用程序考虑使用TSDB，那么新的时间序列数据库Apache IoTDB是您的最佳选择。")]),t._v(" "),v("p",[t._v("发布新版本并完成实验后，我们将更新此页面。")]),t._v(" "),v("p",[t._v("我们也欢迎更多的贡献者更正本文，为IoTDB做出贡献或复现实验。")])],1)}),[],!1,null,null,null);e.default=a.exports}}]);