blktrace btt结果分析
对于btt的结果分析十分的困难，我和同事花了很多的时间在网上查找btt输出的每一项参数的意义，试图更好的分析bio的统计信息，但网上文章一大抄，翻来覆去就是那几篇文章。
本文中内容参考了以下网址：
1.btt官方网址：http://git.kernel.dk/cgit/blktrace/tree/btt/doc/btt.tex
在(centos7.4)/usr/share/doc/blktrace-1.0.5/README中32行提到http://git.kernel.dk/
本文中对btt的分析很多参考自btt.tex
2.http或者git上的blktrace最新版本
http://brick.kernel.dk/snaps/
dd if=/dev/zero of=/dev/sdb bs=1M count=200
mount -t debugfs none /sys/kernel/debug/
blktrace -d /dev/sdb
blkparse -i sdb -d sdb.blktrace.bin
btt -i sdb.blktrace.bin
conv=fdatasync
conv=fsync
conv=sync
oflag=dsyn
oflag=syn
oflag=direct
dd默认是buffer io, 页面是pdflush刷的。自己加上去这些参数看看分析结果，会让你大吃一惊的。
btt结果如下：
btt -i sdb.blktrace.bin | grep -A 100 "All Devices"
==================== All Devices ====================
ALL MIN AVG MAX N
--------------- ------------- ------------- ------------- -----------
Q2Q 0.000000479 0.007676882 5.711474239
Q2A 0.000080329 0.000887075 0.029918892
Q2G 0.000000441 0.000296168 0.029764843
S2G 0.014563643 0.021455586 0.029762593
G2I 0.000000837 0.001079771 0.017816477
Q2M 0.000000134 0.000000299 0.000001491
I2D 0.000019345 0.055941845 0.106661743
M2D 0.000029110 0.000044429 0.000053098
D2C 0.000176810 0.023186230 0.053523069
Q2C 0.000327039 0.079803764 0.146987975
==================== Device Overhead ====================
DEV | Q2G G2I Q2M I2D D2C
---------- | --------- --------- --------- --------- ---------
( , ) | 0.3666% 1.3365% 0.0000% 69.2422% 29.0541%
---------- | --------- --------- --------- --------- ---------
Overall | 0.3666% 1.3365% 0.0000% 69.2422% 29.0541%
==================== Device Merge Information ====================
DEV | #Q #D Ratio | BLKmin BLKavg BLKmax Total
---------- | -------- -------- ------- | -------- -------- -------- --------
( , ) | 1.0 |
==================== Device Q2Q Seek Information ====================
DEV | NSEEKS MEAN MEDIAN | MODE
---------- | --------------- --------------- --------------- | ---------------
( , ) | 1667079.8 | ()
---------- | --------------- --------------- --------------- | ---------------
Overall | NSEEKS MEAN MEDIAN | MODE
Average | 1667079.8 | ()
==================== Device D2D Seek Information ====================
DEV | NSEEKS MEAN MEDIAN | MODE
---------- | --------------- --------------- --------------- | ---------------
( , ) | 1687714.7 | ()
---------- | --------------- --------------- --------------- | ---------------
Overall | NSEEKS MEAN MEDIAN | MODE
Average | 1687714.7 | ()
==================== Plug Information ====================
DEV | # Plugs # Timer Us | % Time Q Plugged
---------- | ---------- ---------- | ----------------
( , ) | ( ) | 1.348107347%
DEV | IOs/Unp IOs/Unp(to)
---------- | ---------- ----------
( , ) | 0.0 0.0
( , ) | 14.8 10.7
---------- | ---------- ----------
Overall | IOs/Unp IOs/Unp(to)
Average | 14.8 10.7
==================== Active Requests At Q Information ====================
DEV | Avg Reqs @ Q
---------- | -------------
( , ) | 71.0
==================== I/O Active Period Information ====================
DEV | # Live Avg. Act Avg. !Act % Live
---------- | ---------- ------------- ------------- ------
( , ) | 0.000000000 0.000000000 0.00
( , ) | 0.134758122 1.351199997 10.09
---------- | ---------- ------------- ------------- ------
Total Sys | 0.134758122 1.351199997 10.09
# Total System
# Total System : q activity
0.000001613 0.0
0.000001613 0.4
0.124339757 0.4
0.124339757 0.0
0.301083076 0.0
0.301083076 0.4
0.400630592 0.4
0.400630592 0.0
0.573512380 0.0
0.573512380 0.4
0.680941099 0.4
0.680941099 0.0
0.855548103 0.0
0.855548103 0.4
0.954668973 0.4
0.954668973 0.0
4.983438230 0.0
4.983438230 0.4
4.983900171 0.4
4.983900171 0.0
5.964745346 0.0
5.964745346 0.4
6.210325030 0.4
6.210325030 0.0
11.921799269 0.0
11.921799269 0.4
11.922199421 0.4
11.922199421 0.0
# Total System : c activity
0.002123105 0.5
0.002123105 0.9
1.057812876 0.9
1.057812876 0.5
4.983851980 0.5
4.983851980 0.9
5.080603626 0.9
5.080603626 0.5
5.966766526 0.5
5.966766526 0.9
6.311149346 0.9
6.311149346 0.5
11.922139935 0.5
11.922139935 0.9
11.922139935 0.9
11.922139935 0.5
12.022423074 0.5
12.022423074 0.9
12.022423074 0.9
12.022423074 0.5
# Per device
# , : q activity
0.000001613 1.0
0.000001613 1.4
0.124339757 1.4
0.124339757 1.0
0.301083076 1.0
0.301083076 1.4
0.400630592 1.4
0.400630592 1.0
0.573512380 1.0
0.573512380 1.4
0.680941099 1.4
0.680941099 1.0
0.855548103 1.0
0.855548103 1.4
0.954668973 1.4
0.954668973 1.0
4.983438230 1.0
4.983438230 1.4
4.983900171 1.4
4.983900171 1.0
5.964745346 1.0
5.964745346 1.4
6.210325030 1.4
6.210325030 1.0
11.921799269 1.0
11.921799269 1.4
11.922199421 1.4
11.922199421 1.0
# , : c activity
0.002123105 1.5
0.002123105 1.9
1.057812876 1.9
1.057812876 1.5
4.983851980 1.5
4.983851980 1.9
5.080603626 1.9
5.080603626 1.5
5.966766526 1.5
5.966766526 1.9
6.311149346 1.9
6.311149346 1.5
11.922139935 1.5
11.922139935 1.9
11.922139935 1.9
11.922139935 1.5
12.022423074 1.5
12.022423074 1.9
12.022423074 1.9
12.022423074 1.5
# Per process
# blktrace : q activity
# blktrace : c activity
0.635480135 2.5
0.635480135 2.9
0.719490098 2.9
0.719490098 2.5
0.923163074 2.5
0.923163074 2.9
0.923163074 2.9
0.923163074 2.5
# dd : q activity
# dd : c activity
0.644083682 3.5
0.644083682 3.9
0.656542273 3.9
0.656542273 3.5
0.878359453 3.5
0.878359453 3.9
0.901143153 3.9
0.901143153 3.5
# jbd2 : q activity
4.983438230 4.0
4.983438230 4.4
4.983900171 4.4
4.983900171 4.0
11.921799269 4.0
11.921799269 4.4
11.922199421 4.4
11.922199421 4.0
# jbd2 : c activity
# ksoftirqd : q activity
# ksoftirqd : c activity
0.113965122 5.5
0.113965122 5.9
0.198731512 5.9
0.198731512 5.5
0.307666215 5.5
0.307666215 5.9
0.409794955 5.9
0.409794955 5.5
0.616305736 5.5
0.616305736 5.9
0.676618319 5.9
0.676618319 5.5
0.912045716 5.5
0.912045716 5.9
0.952781433 5.9
0.952781433 5.5
6.020433404 5.5
6.020433404 5.9
6.061857376 5.9
6.061857376 5.5
# kworker : q activity
0.000001613 6.0
0.000001613 6.4
0.124339757 6.4
0.124339757 6.0
0.301083076 6.0
0.301083076 6.4
0.400630592 6.4
0.400630592 6.0
0.573512380 6.0
0.573512380 6.4
0.680941099 6.4
0.680941099 6.0
0.855548103 6.0
0.855548103 6.4
0.954668973 6.4
0.954668973 6.0
5.964745346 6.0
5.964745346 6.4
6.210325030 6.4
6.210325030 6.0
# kworker : c activity
0.002123105 6.5
0.002123105 6.9
0.123954697 6.9
0.123954697 6.5
0.303307770 6.5
0.303307770 6.9
0.398885246 6.9
0.398885246 6.5
0.578128551 6.5
0.578128551 6.9
0.680640684 6.9
0.680640684 6.5
0.857973954 6.5
0.857973954 6.9
0.866998251 6.9
0.866998251 6.5
5.966766526 6.5
5.966766526 6.9
6.189072144 6.9
6.189072144 6.5
# pid000000000 : q activity
# pid000000000 : c activity
0.028265966 7.5
0.028265966 7.9
1.057812876 7.9
1.057812876 7.5
4.983851980 7.5
4.983851980 7.9
5.080603626 7.9
5.080603626 7.5
6.014711621 7.5
6.014711621 7.9
6.311149346 7.9
6.311149346 7.5
11.922139935 7.5
11.922139935 7.9
11.922139935 7.9
11.922139935 7.5
12.022423074 7.5
12.022423074 7.9
12.022423074 7.9
12.022423074 7.5
# rcu_sched : q activity
# rcu_sched : c activity
0.916222959 8.5
0.916222959 8.9
0.916659526 8.9
0.916659526 8.5
每一段详细分析：
ALL MIN AVG MAX N
--------------- ------------- ------------- ------------- ----------- Q2Q 0.000000479 0.007676882 5.711474239
Q2A 0.000080329 0.000887075 0.029918892
Q2G 0.000000441 0.000296168 0.029764843 1535
S2G 0.014563643 0.021455586 0.029762593
G2I 0.000000837 0.001079771 0.017816477 1535
Q2M 0.000000134 0.000000299 0.000001491
I2D 0.000019345 0.055941845 0.106661743 1535
M2D 0.000029110 0.000044429 0.000053098
D2C 0.000176810 0.023186230 0.053523069 1554
Q2C 0.000327039 0.079803764 0.146987975 1554
我们需要关注的是红色部分，单位是秒s，N表示个数。
==================== Device Overhead ====================
DEV | Q2G G2I Q2M I2D D2C
---------- | --------- --------- --------- --------- ---------
( , ) | 0.3666% 1.3365% 0.0000% 69.2422% 29.0541%
---------- | --------- --------- --------- --------- ---------
Overall | 0.3666% 1.3365% 0.0000% 69.2422% 29.0541%
在此次测试IO中，I2D(IO请求在IO调度中所耗费的时间)平均花费了69%的时间，属于最耗时的部分，其次是D2C(IO请求在硬件设备中耗费的时间)花费了29%的时间。
==================== Device Merge Information ====================
DEV | #Q #D Ratio | BLKmin BLKavg BLKmax Total
---------- | -------- -------- ------- | -------- -------- -------- --------
( , ) | 1.0 |
对此部分的解释如下：
A key measurement when making changes in the system (software \emph{or} hardware) is to understand the block IO layer ends up merging incoming requests into fewer, but larger, IOs to the underlying driver. In this section, we show the number of incoming requests (Q), the number of issued requests (D) and the resultant ratio. We also provide values for the minimum, average and maximum IOs generated.
IO合并请求信息，Q表示传入的IO请求，D表示合并后发出的请求，D越小证明数据包越大，合并请求比例越高(越高越好)。
==================== Device Q2Q Seek Information ====================
DEV | NSEEKS MEAN MEDIAN | MODE
---------- | --------------- --------------- --------------- | ---------------
( , ) | 1667079.8 | ()
---------- | --------------- --------------- --------------- | ---------------
Overall | NSEEKS MEAN MEDIAN | MODE
Average | 1667079.8 | ()
==================== Device D2D Seek Information ====================
DEV | NSEEKS MEAN MEDIAN | MODE
---------- | --------------- --------------- --------------- | ---------------
( , ) | 1687714.7 | ()
---------- | --------------- --------------- --------------- | ---------------
Overall | NSEEKS MEAN MEDIAN | MODE
Average | 1687714.7 | ()
# Total System
# Total System : q activity
# Total System : c activity # Per device
# , : q activity
# , : c activity # Per process
# blktrace : q activity
# blktrace : c activity
# dd : q activity
# dd : c activity
# jbd2 : q activity
# jbd2 : c activity
# ksoftirqd : q activity
# ksoftirqd : c activity
# kworker : q activity
# kworker : c activity
# pid000000000 : q activity
# pid000000000 : c activity
# rcu_sched : q activity
# rcu_sched : c activity
kworker(KWorker is in process ...At the moment there is not active source code to download because the project is in very early version. The only active place on this page is the bug tracking page that keeps my current list of bugs-tracks open. When the project hits the 0.1 release (see bug tracking) it will be available to the public (and maybe to some volunteers).)
jbd2(journaling block driver)这个进程实现的是文件系统的日志功能，磁盘使用日志功能来保证数据的完整性
RCU(Read-Copy Update)
ksoftirqd软中断处理线程
对于activity的定义可以参见http://git.kernel.dk/cgit/blktrace/tree/btt/output.c
部分代码如下：
int output_regions(FILE *ofp, char *header, struct region_info *reg,
float base)
{
if (list_len(®->qranges) == && list_len(®->cranges) == )
return ; fprintf(ofp, "# %16s : q activity\n", header);
__output_ranges(ofp, ®->qranges, base);
fprintf(ofp, "\n"); fprintf(ofp, "# %16s : c activity\n", header);
__output_ranges(ofp, ®->cranges, base + 0.5);
fprintf(ofp, "\n"); return ;
}
本文参考
http://git.kernel.dk/cgit/blktrace/tree/btt/doc/btt.tex
blktrace btt结果分析的更多相关文章
- IO在block级别的过程分析
btt User Guide在百度找了3天没找到,bing也不行,结果google第一页第5个结果就是. 可恶的GFW http://www.fis.unipr.it/doc/blktrace-1.0 ...
- 利用BLKTRACE分析IO性能
在Linux系统上,如果I/O发生性能问题,有没有办法进一步定位故障位置呢?iostat等最常用的工具肯定是指望不上的,[容易被误读的iostat]一文中解释过await表示单个I/O所需的平均时间, ...
- [转] 利用BLKTRACE分析IO性能
在Linux系统上,如果I/O发生性能问题,有没有办法进一步定位故障位置呢?iostat等最常用的工具肯定是指望不上的,[容易被误读的iostat]一文中解释过await表示单个I/O所需的平均时间, ...
- blktrace分析IO
http://bean-li.github.io/blktrace-to-report/ 前言 上篇博客介绍了iostat的一些输出,这篇介绍blktrace这个神器.上一节介绍iostat的时候,我 ...
- [转载]blktrace分析IO
前言 上篇博客介绍了iostat的一些输出,这篇介绍blktrace这个神器.上一节介绍iostat的时候,我们心心念念希望得到块设备处理io的service time,而不是service time ...
- 通过blktrace, debugfs分析磁盘IO
前几天微博上有同学问我磁盘util达到了100%时程序性能下降的问题,由于信息实在有限,我也没有办法帮太大的忙,这篇blog只是想给他列一下在磁盘util很高的时候如何通过blktrace+debug ...
- 【转】通过blktrace, debugfs分析磁盘IO
前几天微博上有同学问我磁盘util达到了100%时程序性能下降的问题,由于信息实在有限,我也没有办法帮太大的忙,这篇blog只是想给他列一下在磁盘util很高的时候如何通过blktrace+debug ...
- Linux服务器I/O性能分析-2
一.如何正确分析IO性能 1.1 BLKTRACE分析IO性能 之前的文章已经说明,要是系统发生I/O性能问题,我们常用的命令是无法精确定位问题(内核I/O调度器消耗的时间和硬件消耗的时间,这个不能作 ...
- 【Android】源码external/目录中在编译过程中生成的文件列表
=> external/eyes-free: accessibilityvalidator.jar (host,share) => external/mesa3d: libMesa ...
随机推荐
- C++中ceil、floor和round的区别
Math类中提供了三个与取整有关的方法:ceil,floor,round,这些方法的作用于它们的英文名称的含义相对应 1.ceil的英文意义是天花板,该方法就表示向上取整,Math.ceil(11.3 ...
- [原]visual studio 将（无扩展名）文件以某种（C++）方式阅读（映射）
工具 选项 文本编辑器 文件扩展名 选择“将无扩展····”后面设置一下就可以
- 递归--练习3--noi7592求最大公约数问题
递归--练习3--noi7592求最大公约数问题 一.心得 两个低级错误:1. ll setMax(ll &m,ll &n)中无引用,结果只传值,没传地址2. return f(n,m ...
- hdu3032sg打表找规律
先打个表冷静一下 #include<map> #include<set> #include<cmath> #include<queue> #includ ...
- Oracle Sourcing Implementation and Administration Guide(转)
原文地址 Oracle Sourcing Implementation and Administration Guide
- Sizzle源码分析：一 设计思路
一.前言 DOM选择器(Sizzle)是jQuery框架中非常重要的一部分,在H5还没有流行起来的时候,jQuery为我们提供了一个简洁,方便,高效的DOM操作模式,成为那个时代的经典.虽然现在Vue ...
- facade外观模式
通过买股票与通过基金买股票引出外观模式: package com.disign.facade; /** * Created by zhen on 2017-05-18. */ public class ...
- L175 Endorestiform Nucleus: Scientist Just Discovered a New Part of the Human Brain
The newly named Endorestiform Nucleus sits in the inferior cerebellar小脑 peduncle, at the junction be ...
- YAML 与 front-matter
1. YAML 类似 Linux:Linux is not UniX,YAML:YAML ain't markup language,是一种递归缩写,是一个可读性高并且容易被人类阅读,容易和脚本语言交 ...
- 使用 某款基于Socks5协议的代理软件 一段时间后 被封锁掉IP的一些技术思考
由于关键词比较敏感为了不被删除帖子所以文中某软件(上图所示软件)不用全称表示. 去年9月末在 在某国外网站 上弄了一个vpn,在上面安装了某软件,使用起来还是蛮不错的,平时查查英文论文,看看美剧还是比 ...