Difference between revisions of "Benchmarks"

From Linuxintro
imported>ThorstenStaerk
imported>ThorstenStaerk
(there are two different netcat versions, this one is common on SUSE now)
 
(16 intermediate revisions by 5 users not shown)
Line 1: Line 1:
= disk =
+
A benchmark is everything that gives you an idea about performance. For your Linux, you can use the following benchmarks:
 +
 
 +
= Combined benchmarks =
 +
A quick and easy benchmark that combines the performance of all your computer's components is [http://www.primatelabs.ca/geekbench/ GeekBench].
 +
 
 +
= Component benchmarks =
 +
assalamu aleykum!  u menya diketsy artrit, ya chitala chto ee mojno vilechit s hijamoy.No vrechi i mirskaya medecina govorit chto moya bolezn ne izlechima. Ne mogli bi vi rasskazat nemnogo podrobney kak vilechit abЮвенильный артритbb ?  Da vozdast vam Allah blagom
 +
 
 +
== RAM ==
 +
Quick and dirty test if your RAM is okay:
 +
# dd if=/dev/zero of=/dev/shm/test count=5000 bs=8k
 +
5000+0 records in
 +
5000+0 records out
 +
40960000 bytes (41 MB) copied, 0.0175427 s, 2.3 GB/s
 +
 
 +
== disk ==
 +
When doing an I/O benchmark you need to keep track of the following:
 +
* blocksize/cache/random/IOPS
 +
* read/write
 +
* latency
 +
* processor utilization
 +
 
 +
=== block device ===
 
* hdparm
 
* hdparm
 +
* [http://benjamin-schweizer.de/files/iops/ IOps] - random read performance in IOPS and MB/s
  
== file system ==
+
=== file system ===
 
* bonnie
 
* bonnie
 
* iozone
 
* iozone
 +
* [[iometer]]
 +
 +
== networking ==
  
= networking =
+
=== Throughput ===
  
== latency ==
+
Use [[netcat]], for example on the receiver:
 +
netcat -l 8000 >/dev/null
 +
and on the sender in the [[bash]]:
 +
dd if=test bs=1024K count=512 > /dev/tcp/192.168.0.9/8000
 +
a result can be:
 +
4887552 bytes (4.9 MB) copied, 4.3689 s, 1.1 MB/s
 +
 
 +
An interesting thing is that if you forget the ''>/dev/null'' you will come to a pretty constant value which is useless because it tells you how quickly the [[shell]] can write nulls.
 +
 
 +
=== latency ===
 
* ping
 
* ping
  
== httpd benchmarks ==
+
=== MTU ===
 +
Increasing the MTU helps reducing the need for ACK packages and reduces interrupt storm on the processors.
 +
 
 +
=== httpd benchmarks ===
 
* [http://www.vanheusden.com/httping/ httping]
 
* [http://www.vanheusden.com/httping/ httping]
 
webserver stressen
 
webserver stressen
Line 21: Line 59:
  
 
== Graphic card ==
 
== Graphic card ==
Watching a [[video]] feels sluggish on one computer, but on the other it is okay ? Test your [[graphics card]] with [[glxgears]] like this:
+
Watching a [[video]] feels sluggish on one computer, but on the other it is okay ? Test your graphics card with [[glxgears]] like this:
 
<pre>
 
<pre>
 
$ glxgears
 
$ glxgears
Line 38: Line 76:
 
320 frames in 5.1 seconds = 62.149 FPS
 
320 frames in 5.1 seconds = 62.149 FPS
 
</pre>
 
</pre>
 +
 +
= See also =
 +
* [[latency]]
 +
* [[block sizes]]
 +
* [http://www.howtoforge.com/how-to-benchmark-your-system-cpu-file-io-mysql-with-sysbench SysBench]

Latest revision as of 10:13, 26 January 2016

A benchmark is everything that gives you an idea about performance. For your Linux, you can use the following benchmarks:

Combined benchmarks

A quick and easy benchmark that combines the performance of all your computer's components is GeekBench.

Component benchmarks

assalamu aleykum! u menya diketsy artrit, ya chitala chto ee mojno vilechit s hijamoy.No vrechi i mirskaya medecina govorit chto moya bolezn ne izlechima. Ne mogli bi vi rasskazat nemnogo podrobney kak vilechit abЮвенильный артритbb ? Da vozdast vam Allah blagom

RAM

Quick and dirty test if your RAM is okay:

# dd if=/dev/zero of=/dev/shm/test count=5000 bs=8k
5000+0 records in
5000+0 records out
40960000 bytes (41 MB) copied, 0.0175427 s, 2.3 GB/s

disk

When doing an I/O benchmark you need to keep track of the following:

  • blocksize/cache/random/IOPS
  • read/write
  • latency
  • processor utilization

block device

  • hdparm
  • IOps - random read performance in IOPS and MB/s

file system

networking

Throughput

Use netcat, for example on the receiver:

netcat -l 8000 >/dev/null

and on the sender in the bash:

dd if=test bs=1024K count=512 > /dev/tcp/192.168.0.9/8000 

a result can be:

4887552 bytes (4.9 MB) copied, 4.3689 s, 1.1 MB/s

An interesting thing is that if you forget the >/dev/null you will come to a pretty constant value which is useless because it tells you how quickly the shell can write nulls.

latency

  • ping

MTU

Increasing the MTU helps reducing the need for ACK packages and reduces interrupt storm on the processors.

httpd benchmarks

webserver stressen

httping -f -h servername
httping -g URL

HTTP benchmarking utility

Graphic card

Watching a video feels sluggish on one computer, but on the other it is okay ? Test your graphics card with glxgears like this:

$ glxgears
359 frames in 5.1 seconds = 70.397 FPS
320 frames in 5.1 seconds = 62.590 FPS
320 frames in 5.3 seconds = 60.201 FPS
320 frames in 5.1 seconds = 63.046 FPS
300 frames in 5.4 seconds = 55.305 FPS
340 frames in 5.4 seconds = 62.724 FPS
300 frames in 5.2 seconds = 57.543 FPS
340 frames in 5.4 seconds = 62.483 FPS
320 frames in 5.1 seconds = 62.149 FPS
320 frames in 5.2 seconds = 61.365 FPS
300 frames in 5.1 seconds = 59.275 FPS
320 frames in 5.5 seconds = 58.109 FPS
320 frames in 5.1 seconds = 62.149 FPS

See also