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(&reg->qranges) == && list_len(&reg->cranges) == )
return ; fprintf(ofp, "# %16s : q activity\n", header);
__output_ranges(ofp, &reg->qranges, base);
fprintf(ofp, "\n"); fprintf(ofp, "# %16s : c activity\n", header);
__output_ranges(ofp, &reg->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

  1. 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 ...

  2. 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, ...

  3. [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, ...

  4. 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 ...

  5. [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 ...

  6. 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 ...

  7. 【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 ...

  8. 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 ...

  9. 【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

  1. Android之shape与selector实现圆角

    shape和selector是Android UI设计中经常用到的,比如我们要自定义一个圆角Button,点击Button有些效果的变化,就要用到shape和selector.可以这样说,shape和 ...

  2. JAVA消息 JMS 很重要

    首先大致讲一下,java 消息模块 消息,个人理解分为两种:1.同步消息(RPC调用) 2.异步消息(本篇讲解部分) 一.同步消息java提供了多种方案: 最新比较常用的方式就是spring Http ...

  3. hdu 4845 状压bfs(分层思想)

    拯救大兵瑞恩 Time Limit: 3000/1000 MS (Java/Others)    Memory Limit: 65535/32768 K (Java/Others)Total Subm ...

  4. UVA-11419 SAM I AM (最小点覆盖)

    题目大意:在一个n*m的网格中,有k个目标,现在可以任选一行或列消除在其上的所有目标,求出最少选择次数及选法. 题目分析:经典的最小点覆盖问题,并且输出一个最小点覆盖集.在求出最大匹配之后,以未覆盖的 ...

  5. Android数据库升级不丢失数据解决方案

    在Android开发中,sqlite至关重要,增删查改不多说,难点在于,1,并发,多个线程同时操作数据库.2,版本升级时,如果数据库表中新加了个字段,如何在不删除表的情况下顺利过渡,从而不丢失数据. ...

  6. String类的concat()方法

    String类的concat()方法: public class MyClass { public static void main(String[] args) { String str1=&quo ...

  7. Alpha阶段第1周Scrum立会报告+燃尽图 05

    作业要求与https://edu.cnblogs.com/campus/nenu/2018fall/homework/2246相同 一.小组介绍 组长:刘莹莹 组员:朱珅莹 孙韦男 祝玮琦 王玉潘 周 ...

  8. STL标准库-容器-map和multimap

    技术在于交流.沟通,本文为博主原创文章转载请注明出处并保持作品的完整性 map与multimap为关联容器,结构如下 map底层实现依然是rb_tree 他的data可以改,但是key不能改,因此ma ...

  9. I.MX6 eMMC 中启动U-boot存放的首地址

    /************************************************************************************ * I.MX6 eMMC 中 ...

  10. 使用 Windows 10 中的加速度计(Accelerometer,重力传感器)

    在做 UWP 应用开发的时候还有什么理由可以用到加速度计呢?场景很多啦,比如做游戏,做类似 Surface Hub 那种一边旋转,一边所有内容跟着一起转的效果. Windows 10 UWP 中的加速 ...