A poor man's benchmark and stopwatch
Contents
Recently I wanted to measure how long it would take my program to execute a certain action. I didn't want to search for time-related functions in that language, so I had just been going to just use my Casio watch. However, I suddenly realized that I forgot them at home...
"Wait a minute", I thought, "I have a PC with a modern operating system, openSuSE Linux! How come it doesn't have a stopwatch program?" I scratched my head, but couldn't remember anything like that in standard Linux utils.
I asked my peers about such a program, and Konstantin Vlasov proposed a solution: use... dd!
What is dd?
If you still don't get it, I'll remind you. dd is a standard Unix (Hence Linux) tool... to perform low-level data stream copying. Its most popular use is to clone raw partitions (as array of bytes, not as sets of files):
dd if=/dev/sdb5 of=/dev/sdc1 dd if=/dev/cdrom of=image.iso
But we'll get to its conventional use later, and now...
dd as stopwatch
The funny thing about dd is that in the default mode it prints the time it's been running. Indeed, dd is a data copying program, so it's natural to print, at the end, how much data it has copied, and how fast.
So it prints total size of the data copied, the time it took it to do this, and the resultant speed... Wait a minute... the time! That's it!
To use dd as stopwatch, you don't need to copy actual data. Just run it without arguments and hit Ctrl+C when you need to stop the countdown. Here's how the output will look like:
pavel@lonely ~ $ dd ^C0+0 records in 0+0 records out 0 bytes (0 B) copied, 5.97049 s, 0.0 kB/s
There it is, the stopwatch triggered by just Enter and Ctrl+C keys.
Note that instead of triggering the stop manually with Ctrl+C, you may just send a signal to dd, and have the same effect. We'll utilize this in the further section.
dd as a benchmark
So, we've just learned that dd can print speed of the data copy operation it performed. And Linux has nice pseudo-files that just generate streams of data... why don't we try to use this files to measure how fast can a computer copy nothing to nowhere?
Let's check how much data your machine will copy in ten seconds! Here's a sample listing:
pavel@lonely ~ $ dd if=/dev/zero of=/dev/null bs=64k & sleep 10 ; kill -INT %1 ; wait %1 [1] 885 1434838+1 records in 1434838+0 records out 94033543168 bytes (94 GB) copied, 10.0106 s, 9.4 GB/s [1]+ Interrupt dd if=/dev/zero of=/dev/null bs=64k
Standard Linux shell, Bash, has quite simple features to control it child processes. For example, the code at the left features "job specifications" ("jobspecs"). When a program is run in the background (with use of & at the end of a statement, instead of the usual ;), it gets a job number, and its pid can be gotten by writing %n, where n is that number.
In our code we feature sending a a signal to a certain jobspec, and waiting for it (if it weren't for wait, the shell prompt could mingle with the output of the dd's signal handler). And if we only have one job running in current shell, its jobspec will always be %1.
This command will launch dd process that copies a stream of zero bytes maintained by kernel (/dev/zero) to stream that chews everything it gets without any effect (/dev/null). You may try to test how this simple copying works and learn how fast is your... well, it's hard to tell, what exactly, but you can compare computers with your friends and beat them in this competition! We've held dd benchmarking comparison at the message board of our university, and the top result was 70.6 GB/s, which is like ten times faster than on my machine!
By the very same means you may analyze performance of your disk, by putting a file as the of= argument. But that's a poor man's solution anyway, as there are better benchmarks both for disks and for memory.
Other (not interesting) uses of dd
Aside from copying raw partitions (it's especially useful if you're trying to recover your disk, and accessing it as a file system requires a lot of unnecessary reads, which may be harmful), dd may also be used to create big files of specific size. Why would one want it?... Right, to create swap files. Here's a sample listing to create a 1 Gb swap file
I learned this in a hard time at work, where an important launch of our experiments had the controlling program leaked 4 Gb of memory...
Here's a (really good) Wikipedia article about various uses of dd that presents even more examples of the neat tricks you can employ dd for.
***
Although I'm an experienced Linux user, it doesn't stop to surprise me. The dd program is not the only tool that, due to its simplicity, serves different purposes. That simplicity of basic tools, which leads to their powerful combinations, is one of the things in Linux I appreciate most.
And sometimes it can be even amusing.
Comments imported from the old website
It's ironic that I didn't think about time program to measure time. Perhaps, that's a matter of native language, when I don't have an entanglement between the words "time" and "время".
Anyway, to make time wait for your reaction and to measure, you should specify a program that would wait for your input. You might use time grep x or time cat, but the shortest choice would anyway be time dd, which makes time part unnecessary. :-)
And of course, as the smiley you finished your message with alleges, precision doesn't matter. Human is quite slow to respond to events. To notice the change in color of a circle a human needs 0.2s, and the time is bigger for complex events. Instead of linking to research, I'd just suggest you to measure your own reaction on the simpler event, and on the more complex one.
In addition to dd, there is also the GNU time utility, which I think is standard in all Linux distributions. You can find the time used by a program if you use it like "time make", or for example:
Unfortunately, unlike dd, time only is accurate to milliseconds :)