🚀 go-pugleaf

Paul Miach writes:
> herm@coombs.anu.edu.au (Herman H. Fredriksson) writes:
> 
>>The ping times to USA from Australia have been greatly degraded for
>>the last few days (week?), an example made from coombs.anu.edu.au
> 
>>----arc2.nsn.nasa.gov PING Statistics----
>>22 packets transmitted, 21 packets received, 4% packet loss
>>round-trip (ms)  min/avg/max = 1333/1616/2016 
> 
>>Is this a temporary thing? Any info? (have I missed something? :)

nope - we are stuck with this for quite some time yet (although things
may improve over the Christmas week!).

> 	Two weeks ago, when I last looked (via snmp) the usage of the link
> comming from the USA was 100%, so it is probably queing delays making things
> slow. Mean while, the link in the other direction was at about 50%. Don't
> discount solar effects either, there was a post a little while ago, pitty I
> can't remember the dates.

Observed  ping  times  are  slow  and  highly  variable  because  the  link is
completely saturated and you are now getting into  the  practice  of  queueing
theory.  Queueing  theory  in  this  context  states  that  the variability of
roundtrip times increases exponentially once you drive  the  resource  into  a
mode where queueing starts to occur on the gateways. 

You should also note with considerable concern the packet loss going non-zero. 
This is the practice of what happens after queueing theory fails due to buffer 
space  exhaustion  and  indicates  that  not  only is the link sitting at 100% 
utilisation but that the packet arrival rate at the device is in excess of the 
link capacity for extended periods of time. 

Now  one  could  say  that an improvement in capacity could push this queueing
back from practice to theory - and you'd be right. However this is not a cheap 
option  and  right  now   a   position   of   budgtary   constraint   at   the 
Vice-Chancellorial  level  implies  that we may have to live with this problem 
for some time yet. Once there is positive news to  send  out  about  regarding 
upgrade  dates  and  capacity  you can be sure we will post it out - right now 
there's nothing to say. 

> 	If anyone can have a guess, how effective has archie.au been in
> reducing usage on the link?

There  is  a  lot  of  traffic on archie and it has made a difference (we have
stats on its use and on that basis I can  indicate  that  it  has  worked  and
continues  to work highly effectively) - the difference is that the problem of
complete saturation and consequent degradation of the link to the rest of  the
Internet has become critical now rather a month or two ago. In bringing such a 
pervasively  useful  facility  into a highly receptive environment exponential 
growth has been a continual feature of the network's growth. Managing  such  a 
pattern  of  growth  is  a  major challenge - both technically and in terms of 
generating the finances to  support  the  every  increasing  requirements  for 
capacity. 

(btw  don't sit there and verify the problem with more pings across the link -
needless to say it simply adds to the problem!) 


-- 
Geoff Huston,                           	e-mail: G.Huston@aarnet.edu.au
Manager,	 				phone: +61 6 249 3542
Australian Academic Research Network		fax:   +61 6 249 1369