A guide on how to diagnose and fix a sluggish Linux machine.

1. Help! My Computer is Sluggish
Kevin B. O'Brien
Washtenaw Linux Users Group

2. Just buy a new CPU!
 This is what most people will try
 It is actually a less likely way to solve your
problem
 You need to diagnose the actual cause before
you know what to do

3. Some Suggestions
 This presentation is not exhaustive
 Linux has many, many tools for analyzing
computer problems
 So we will cover just a few of the things you can
try
 Read the man pages for the tools we cover.
There is a lot of added information there that
can help you do even more with the tools.

4. Hardware vs. Software
 Hardware = the part of a computer you can hit
 Software = the part you can't hit
 A computer is always a combination of both
hardware and software
 You should look at both if you are having
problems

5. Software
 Software can be buggy, badly written, or
corrupted in some way
 If it is buggy or badly written, you can look for
alternatives
 If it is corrupted you can do a reinstallation

6. Top 1
 This is a very useful tool to see what your
software is doing
 Just open a terminal and type “top”

7. Top 2

8. Top 3
 This command lists all of the running processes
on your computer
 It also shows how much CPU and memory
each one is using
 You can sort the display by pressing an upper-
case letter “O”
 For instance, the sequence upper-case “O”
-->n-->”enter” will sort by Memory used

9. Top 4

10. Top 5
 You want to look for processes that use a lot of
CPU or Memory as potential problems
 Large numbers are not always a problem,
though. I have a Virtual Box VM open running a
Windows VM. That is an expected load that I
took into account.
 But if Firefox is using 25% of your memory or
CPU, that is probably a problem

11. The kill test
 One way you can test if a particular app is the
problem is to close it
 Sometimes the app is not responding, so you
need to kill it
 With top open in the terminal, type “k”
 You will get an added prompt above the list of
processes asking which one you want to kill
 Just enter the PID (process ID) of the process

12. KDE
 If you use KDE you can bring up a system
monitor from krunner
 Press Alt+F2 to bring up krunner
 Click the second icon from the left, which looks
like a small graph

13. krunner

14. Gnome
 Go to System → Administration → System
Monitor

15. Gnome System Monitor

16. Note on Multi-Core Processors
 Each process runs on a single core
 Top, and the Gnome System Monitor report the
percentage of that single core that each
process uses
 KDE's System Activity reports the percentage
of all processor cycles available on all cores, so
it reports a smaller number
 Pay more attention to top, since maxing out a
single core is still a problem even if you have
another core available

17. Interpreting the results
 If one process/application stands out, and
closing/killing it makes the problem go away,
you may have a software problem
 You can then try to reinstall or find alternative
software
 Solving a software problem may be a little work,
but it costs nothing. You should rule out
software problems before looking at a hardware
problem, therefore.

18. Hardware problems
 There are potentially four problem areas:
 CPU
 RAM
 Disk I/O
 Video Card
 Knowing the source of the problem will let you
resolve it without wasting money on unneeded
hardware upgrades

19. Where people go wrong
 Many people will assume a sluggish computer
needs a CPU upgrade
 More knowledgeable folks know that more RAM
is a lot more likely to work than a new CPU
 In a PC World study (October 2010) RAM also
beat out a faster hard drive for improving
performance.

20. Uptime

21. Load 1
 Look at the load figures
 They show the load over the past 1, 5, and 15
minutes
 Load is the average number of processes that
have to wait for CPU time
 You adjust for the number of cores. A load of 2
on a two-core system is equivalent to a load of
1 on a single-core system.

22. Load 2
 Load is just one indication
 You can get even more from the top command
 Notice that the first line of the top results is the
same as you get from uptime

23. Top, again

24. CPU Load 1
 Look at the third line
 Us is the percentage of CPU time taken up by
user processes. Note that user is not just a
person, it could be Apache, MySQL, etc. If this
is very high, it may be an indication of a high
CPU load.
 Sy is the percentage used by the kernel and
other system processes

25. CPU Load 2
 Id is the idle time. The higher the better here. If
this is reasonably high you can be pretty sure
you don't have a CPU problem.
 Wa is the I/O wait time. This indicates the
percentage of time that the CPU was waiting for
I/O, which is usually your disk. A high number
here indicates a disk I/O problem, and might be
solved by getting a faster drive

For the others, try man top for an explanation

26. CPU Load 3

If you see a high number for us or sy, look at
the processes involved, like we did for software
 You might see a high number related to a
particular application. That application may be
something you can fix or replace. But if it is
proper, like a high figure for MySQL on a
database or web server, you may need a better
CPU.
 If it is a large number of separate processes
with no one being large, that is also an
indication that a new CPU is required

27. RAM
 This is in fact the most common problem area,
and more RAM will often clear up any problem
you have
 RAM is the workspace for the computer. All
code has to be copied into RAM before the
CPU can work with it.
 When RAM runs out, code that may be less
frequently used gets copied to “swap”. This is
an area of reserved space on a disk.

28. Swap and RAM Performance
 Because disk I/O is much slower than RAM,
code that is moved to swap is slower to access
 If RAM is inadequate, code gets moved in and
out of the swap area constantly, causing “disk
thrashing”. This refers to constant hard drive
activity, which can often be heard or seen as a
flashing LED.

29. RAM Utilization
 Do not worry if your RAM appears to be
completely utilized
 This is not like a high percentage of CPU
utilization
 RAM is supposed to be used, that is why it is
there
 For instance, shared libraries are frequently
kept in RAM, which can speed up the launch of
other programs that use those libraries

30. RAM as Cache 1
 To understand this, think of the RAM as a kind
of cache
 Files in cache are always faster to access than
files on the hard drive
 So the kernel will keep anything you have used
in RAM in case it is needed again
 Any system that is used for a few hours will
almost certainly see all of the RAM in use
therefore

31. RAM as Cache 2
 Because a lot of your RAM is being used as a
type of cache, that part of the RAM will be
released as needed
 So if you start a new process and need more
RAM for it, some of the cached code is cleared
and the RAM is used for the new process

32. RAM numbers in top

33. Interpreting RAM numbers

Note that line 4 is Mem
 This shows that I have 16 GB of RAM
 And almost all of it is used. Only 113MB is free
 Is this a problem?

No, look at the next line for Swap

34. Swap in top

Look at line 5, Swap
 There you see I have 6GB of swap space, but
only 1.7MB is being used. Almost all of my
swap space is empty.
 Look at the cache number. This indicates how
much RAM is being used as a cache.
 To see what is available here, look at cache
and free swap. In this case, there is 8GB of
cache, and 6GB of swap. No RAM problems
here!

35. Check RAM first
 You want to check Swap issues before
checking other I/O issues because a RAM
problem can look like a disk I/O problem. This
is because the Swap is relying on disk I/O.
 If you do have a RAM issue, look at the
processes to see which one(s) are responsible
 Remember you can sort by memory used by
the sequence upper-case “O”
-->n-->”enter”

36. High RAM usage
 Look for a process that uses an abnormally
high amount of RAM
 This could be an indication of a memory leak
 This is when an application does not release
memory when it is done, so that it gradually
uses up most of the available RAM
 If you find this, look for an update that fixes it,
or look for an alternative
 And if you find this, file a bug!

37. Disk I/O

The tool you want here is called iostat
 This tool is not automatically installed, so you
may need to install it
 You will probably find it in a package called
sysstat
 Install this using your package manager

38. iostat

39. Iostat, the numbers
 On the first line, you can see that I have a 2-
core CPU, I am using a 64-bit version of Linux,
and what kernel I am using

Then we get the same CPU numbers as the top
command gave us
 The bottom section will give you the raw
numbers for disk I/O

40. The I/O numbers
 Tps = transactions per second
 Blk_read/s = blocks read per second
 Blk_wrtn/s = blocks written per second
 Blk_read = total blocks read
 Blk-wrtn = total blocks written

41. Interpretation
 Which drive is getting the activity? Does that
make sense?
 In this case sda is getting most of the action
 This is where my root partition is located, so all
of my kernel and system activity should be here
 Sdb has my home partition. That gets less
activity since most of the time it is sitting idle.

42. Read vs. write
 This can be a clue as well
 Certain applications or processes can be
expected to require more reading or more
writing
 Generally you should expect to have more
reads than writes
 If you have your /var in its own partition, you
should see higher write numbers there than on
other partitions because of the logs being
written

43. iotop

44. Interpreting iotop
 Here, instead of processes, you are looking at
threads
 This could help uncover a thread that is
generating a lot of I/O activity
 In this example, KDE's plasma desktop was on
top, and was writing to disk at 30 K/s
 Does any thread consistently show up in the
top? Does this make sense?

45. Signs of video issues
 The mouse is lagging
 Try dragging a window across the screen, and
it does not move smoothly
 Tearing
 Slow or constant redraws
 These can be caused by CPU issues as well,
so you want to eliminate that first
 Misbehaving apps can cause this (e.g. Flash
plugins)

46. Summary
 Before you spend money, do a systematic
troubleshoot
 Test in the right order
 Check software problems before hardware
problems
 Always do video last since video artifacts can be
caused by CPU load
 Look for results that point somewhere else, e.g. a
memory leak is really a software problem, not a
RAM problem