On Sun, 3 Feb 2019 at 23:25, webwarrior <reg@webwarrior.ws> wrote:
Ben Coman wrote
On Sat, 2 Feb 2019 at 18:54, ducasse <
stepharo@
> wrote:
...
3. Ticked "Free tier only" filter. + Selected "Amazon Linux 2 AMI (HVM), SSD Volume Type" + Clicked <Review and Launch> (used default t2.micro) + Clicked <Launch> + From the pull-down selected "Create a new key pair", gave it a name and clicked <Download Keypair> saved as "SydneyPharoSpeedTest.pem" + Clicked <Launch Instance> + Clicked <View Instances> noted instance... * IP address: 54.252.136.78 * Zone: ap-southeast-2b * Security Group: Launch Wizard 1
4. On my Windows 10 box, in WSL did... $ cd ~/.ssh # if it doesn't exist, first do... mkdir -m 700 ~/.ssh $ cp /mnt/c/Users/Ben/Downloads/SydneyPharoSpeedTest.pem ~/.ssh $ chmod 400 ~/.ssh/SydneyPharoSpeedTest.pem $ ssh -i ~/.ssh/SydneyPharoSpeedTest.pem ubuntu@54.252.136.78$ cat /etc/os-release ID="amzn" ID_LIKE="centos rhel fedora"
GOOD NETWORK BASELINE TEST... Ignoring any packet loss on poor networks, first testing low bandwidths on a good network $ vi test.sh #!/bin/sh if [ -d out ]; then dirdate=`stat -c %z out | awk '{print $1"-"$2}' ` mv out out.${dirdate} fi mkdir out for RATE in 1000k 500k 200k 100k 50k 20k 10k 5k 2k 1k do echo $RATE /usr/bin/time -f "%e" -o out/time.$RATE \ wget --quiet --limit-rate $RATE https://files.pharo.org/pharo-launcher/1.6/pharo-launcher-1.6.msi -O out/file.$RATE & done
$ sh test.sh monitoring with... $ cat out/time* | sort -n $ ls -lS out
results in following table and graphs... $RATE (kb/s) TIME (s) TIME (min) TIME (hr) 1000 54 1 0.0 500 105 2 0.0 200 259 4 0.1 100 515 9 0.1 50 1029 17 0.3 20 2576 43 0.7 10 5149 86 1.4 5 10527 175 2.9
[image: download-speed.png]
Wow that surprised me. I'm not sure what the behaviour of file servers at low bandwidth should be, but intuitively the above seems odd. In the past troubleshooting seems to have been focused on the cause of slow speeds, but these can occur for many reasons unrelated to the the file server. The above test ignores cause to isolate behaviour at slow speeds.
I forgot my own download speed yesterday (today is okay), but here is another sample... "(in Argentina) it is really slow ... 3.5KB/s ... average 10KB/s". http://forum.world.st/Pharo-Downloads-are-sluggish-td5084963.html
I would hope that download time was near linear with speed all the way down to 1kb/s. Anyone have some sysadmins they can lean on to understand if that is realistic?
The straightness of the line using a log-log axis makes it seem like policy rather than physics. [image: download-speed(log).png]
HTH, cheers -ben
download-speed.png (39K) <http://forum.world.st/attachment/5094677/0/download-speed.png>; download-speed(log).png (34K) < http://forum.world.st/attachment/5094677/1/download-speed%28log%29.png>;
Why does that surprise you?
Download time [s] = amount of data [B] / bandwidth [B/s]. Your data and graphs look exactly like they should. When bandwidth -> 0, time -> +infinity.
Hey! Thanks for that reality check. 100kb ==> 515s 10kb ==> 5149s duh! makes perfect sense. I was sure I had checked that simple scaling and it had been out by another magnitude, but I must have had a screw loose from being too caught up in the graph. It was just meant to be a prelude baseline before playing with random packet loss. Sorry for the noise. cheers -ben