packages.macports.org ping time

[Daniel J. Luke]

On Jan 2, 2013, at 12:05 PM, Jeremy Lavergne <jeremy at lavergne.gotdns.org> wrote:
>>> Any US east coast devs: does the French mirror get used often for your archive downloads?
>>> 
>>> From Boston, MA lil.fr.packages.macports.org responds faster to pings than packages.macports.org. Since MacPorts uses ping time to calculate where to make its initial requests from, we might be sending people across the ocean when it isn't for any actual benefit.
>> 
>> how do you know it isn't for any actual benefit? (tcp likes less delay...)
>> 
>>> After spotting the french mirror being used, my initial check saw these times (ping, 5 samples each):
>>> average lil.fr.macports.org response: 102.6ms
>>> average packages.macports.org response: 123.8ms
>> 
>> it would probably be worth comparing file transfer speeds from your location to the two mirrors when you're seeing different ping times like that...
> 
> Alrighty, from the numbers: the french mirror transferred data slower but responded to pings faster.
> 
> $ curl -o /dev/null http://packages.macports.org/qt4-mac/qt4-mac-4.7.3_0+quartz+universal.darwin_10.i386-x86_64.tbz2
>  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
>                                 Dload  Upload   Total   Spent    Left  Speed
> 100  201M  100  201M    0     0   802k      0  0:04:17  0:04:17 --:--:-- 1050k
> 
> $ curl -o /dev/null http://lil.fr.packages.macports.org/qt4-mac/qt4-mac-4.7.3_0+quartz+universal.darwin_10.i386-x86_64.tbz2
>  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
>                                 Dload  Upload   Total   Spent    Left  Speed
> 100  201M  100  201M    0     0   559k      0  0:06:09  0:06:09 --:--:--  560k


That's interesting. From your traceroute it looked like the telehouse IX connection between rcn and tinet might be near capacity (and causing additional latency on that path for you) and so you would perhaps get worse transfer rates via that connection.

It's somewhat difficult to debug things on other people's networks, though ;-)

While you were downloading, did you see initial speed faster to lil.fr but packages.macports.org eventually caught up with/surpassed it?

I'm not sure what any possible solution would be, though. Most (all?) of the mirror servers have good bw availability, so the determining factor for transfer rates would be the delay - and the ping results are usually a good proxy for how 'network close' you are to one of them.

Aside from trying to build our own (crappy) CDN, I'm not aware of a better way to handle this.
--
Daniel J. Luke                                                                   
+========================================================+                        
| *---------------- dluke at geeklair.net ----------------* |                          
| *-------------- http://www.geeklair.net -------------* |                          
+========================================================+                        
|   Opinions expressed are mine and do not necessarily   |                          
|          reflect the opinions of my employer.          |                          
+========================================================+