Marek Goldmann

Docker daemon disk statistics while doing squashing. ``` Linux 3.10.0-327.10.1.el7.x86_64 (performance) 03/23/2016 _x86_64_ (4 CPU) 01:52:41 PM UID PID kB_rd/s kB_wr/s kB_ccwr/s Command 01:52:42 PM 0 25122 0.00 0.00 0.00...

Docker daemon CPU statistics while doing squashing ``` 01:54:00 PM UID PID %usr %system %guest %CPU CPU Command 01:54:01 PM 0 25122 0.00 0.00 0.00 0.00 1 docker 01:54:02 PM...

Now, a bit of CPU stats for the squashing tool itself: ``` 02:16:20 PM 0 26931 0.00 0.00 0.00 0.00 2 python 02:16:21 PM 0 26931 1.00 0.00 0.00 1.00...

And disk usage stats: ``` 02:18:38 PM 0 27021 0.00 4.00 4.00 python 02:18:39 PM 0 27021 0.00 0.00 0.00 python 02:18:40 PM 0 27021 0.00 0.00 0.00 python 02:18:41...

 A lot of reading in the beginning - Docker prepares the tar archive with the image, so reading all layers. It could be faster (at least twice) than...

Small info: my time will be limited over next days, but I'll update this ticket asap, including new graphs and replies. Please give me a few days.

@twaugh is correct - this is local testing that does not involve any `push` or `pull` calls. I'm not investigating at all possibilities of parallel builds. This is not part...

> So, switching to a newer version of Origin 1.2 (or OSE 3.2), Docker 1.9, and enabling parallel pulls should improve this. Currently, all image layers are pulled in serial...

> Is there any way to get a graph showing iowait time during these periods and whether the iowait is because of network or storage? I'll prepare one today.

> So, this and the CPU time used by the Docker daemon are most likely the result of computing the tarsums of all the layers involved. I'm not sure there...