LFS201 Lab 15.1 Comparing I/O Schedulers.

MarkHeathcote

Hi, I'm not sure of the workflow for getting the lab.iosched.sh script to run for this lab.

Im using Ubuntu 20.4 and the command to use was;

$ sudo ./lab_iosched.sh [# reads/writes (NMAX)] [file size in MB (NMEGS)]

How do I get the file into a place where Linux knows where to find it and run it?

heiko_s

Probably chmod +x lab_iosched.sh will do.

luisviveropena

Note: we worked this in the Office Hours session. Need to do just 'sudo ./lab_iosched.sh' for starting. doing 'less lab_iosched.sh' will provide more information about the parameters.

Regards,
Luis.

coop

Yes, you will have to make it executable probably if you downloaded in place. If you got from the "SOLUTIONS" tar ball it already is executable. Your results will depend on distro and kernel. For example I just ran it on Ubuntu 20.04 and the only IO schedulers were "mq-deadline" and "none". On RHEL 8 you have also "bfq" which is actually the default, at least for a vanilla kernel.

heiko_s

Mark: You mentioned during the office hour that you didn't manage to set sudo permissions for your default user. I sent you a private message with examples for /etc/sudoers and /etc/group configurations.

donquijotedelasnubes

hi @luisviveropena , I fininshed this lab, but I dont understand how to compare the results. Could you please hellp me to interpret the results of running this script? I understand that it is comparing four different I/O scheduling strategies, but the results mean what?. For example, what does this means:

testing IOSCHED = mq-deadline
[mq-deadline] kyber bfq none
0.517   0.010   0.752

The full output from the script on my system is like this:

[javier@fedora Downloads]$ sudo ./lab_iosched.sh 16 300
Doing: 16 parallel read/writes on: 300 MB size files

creating as needed random input files

doing timings of parallel reads

 REAL    USER    SYS

testing IOSCHED = mq-deadline
[mq-deadline] kyber bfq none
0.517   0.010   0.752
testing IOSCHED = kyber
mq-deadline [kyber] bfq none
0.522   0.008   0.645
testing IOSCHED = bfq
mq-deadline kyber [bfq] none
0.522   0.008   1.062
testing IOSCHED = none
[none] mq-deadline kyber bfq 
0.527   0.011   1.072

doing timings of parallel writes

 REAL    USER    SYS

testing IOSCHED = none
[none] mq-deadline kyber bfq 
0.102   0.056   0.217
testing IOSCHED = mq-deadline
[mq-deadline] kyber bfq none
0.091   0.033   0.216
testing IOSCHED = kyber
mq-deadline [kyber] bfq none
0.094   0.043   0.220
testing IOSCHED = bfq
mq-deadline kyber [bfq] none
0.102   0.035   0.140

coop

Reading the lab_iosched.sh script that produces the output would tell you what the numbers are. For example the output for the write test is produced by the line:

time do_write_test

the three numbers are: wallclock (real) time, user (process) time, and system (kernel) time, in seconds. There are actually these three headings in the output. In hyour results there are very few differences, as can be explained by choice of hardware, linux kernel etc, and if you are on a virtual machine you may have trouble seeing the differences as it is a fake environment. On my real workstation I get:

ROOT@c9:/tmp/TEMP>../lab_iosched.sh 
Doing: 8 parallel read/writes on: 100 MB size files

creating as needed random input files
100+0 records in
100+0 records out
104857600 bytes (105 MB, 100 MiB) copied, 0.304284 s, 345 MB/s

doing timings of parallel reads

 REAL    USER    SYS

testing IOSCHED = mq-deadline
[mq-deadline] kyber bfq none
0.158   0.009   0.627
testing IOSCHED = kyber
mq-deadline [kyber] bfq none
0.155   0.005   0.637
testing IOSCHED = bfq
mq-deadline kyber [bfq] none
0.164   0.007   0.645
testing IOSCHED = none
[none] mq-deadline kyber bfq 
0.156   0.005   0.689

doing timings of parallel writes

 REAL    USER    SYS

testing IOSCHED = none
[none] mq-deadline kyber bfq 
24.404   0.011   1.052
testing IOSCHED = mq-deadline
[mq-deadline] kyber bfq none
25.426   0.013   1.116
testing IOSCHED = kyber
mq-deadline [kyber] bfq none
25.788   0.011   0.958
testing IOSCHED = bfq
mq-deadline kyber [bfq] none
23.774   0.010   0.891


Once again not much difference.  Modern hardware such as SSD drives has rendered this kind
of issue mostly moot except on certain kinds of situations and the deadline scheduler is almost always the
best choice.  It is here partly just so you understand that the kernel has to make decsions about what to read and write and when and in what order