nanog mailing list archives

Re: Ping - APAC Region


From: Franck Martin <fmartin () linkedin com>
Date: Sat, 2 Apr 2011 23:12:19 +0000

Also remember, you would be serving Australia only from Australia. if I'm
not mistaken, the Australia backbone is more or less volume based
cahrged...

http://www.aarnet.edu.au/services/aarnet-charging.aspx
"AARNet3 charges are different for Shareholders (Members) and for Non
Shareholders (Associates and Affiliates).
  Billing 
  On
Net and Off Net subscriptions are calculated in October each year, and
invoices must be delivered soon after to allow sufficient time for
customers to pay in advance for the following calendar year.
  For
those invoices not paid in full and in advance, On Net and Off Net
Subscriptions, and Access Charges are invoiced by quarter and in
advance. 
  All Usage charges, including Excess Traffic, are invoiced
retrospectively after each quarter. "

On 4/3/11 9:40 , "Matthew Petach" <mpetach () netflight com> wrote:

On Tue, Mar 29, 2011 at 11:17 AM, Matthew Palmer <mpalmer () hezmatt org>
wrote:
On Tue, Mar 29, 2011 at 06:33:07PM +0100, Robert Lusby wrote:
Looking at hosting some servers in Hong Kong, to serve the APAC
region. Our
client is worried that this may slow things down in their Australia
region,
and are wondering whether hosting the servers in an Australian
data-centre
would be a better option.

Does anyone have any statistics on this?

No formal statistics, just a lot of experience.  You may be unsurprised
to
learn that serving into Australia from outside Australia is slower than
serving from within Australia.  That being said, there's a fair bit less
distance for the light to travel from Hong Kong or anywhere in the
region
than from the US.

Given that the bulk of the population density in Australia is on the
eastern coast near Sydney, and the *only* fiber path going anywhere
near Asia from Sydney does so via Guam, the light path traveled from
Sydney to Guam to La Union (PH)  to Hong Kong isn't appreciably
shorter than the light path from Sydney to Hawaii to the US--which is
covered by roughly 6x as many fiber runs as the Guam pathway, and
is thus somewhat cheaper to get onto--you might as well host on the
west coast of the US as in Hong Kong.  (and *that* was a horrific
run-on sentence!)

If I look at average data for the past five years between Sydney and
Hong Kong, San Jose, Singapore, and Los Angeles, on average it's
better to serve Sydney from Los Angeles than Hong Kong or Singapore:

mpetach@netops:/home/mrtg/public_html/performance> ~/tmp/avgperf.pl AUE
HKI
total daily data files read: 1559
AUE to HKI latency (min/avg/max): 134.216/173.273/1052.158
mpetach@netops:/home/mrtg/public_html/performance> ~/tmp/avgperf.pl AUE
SJC
total daily data files read: 1558
AUE to SJC latency (min/avg/max): 149.829/176.674/308.637
mpetach@netops:/home/mrtg/public_html/performance> ~/tmp/avgperf.pl AUE
SG1
total daily data files read: 1558
AUE to SG1 latency (min/avg/max): 101.871/204.485/999
mpetach@netops:/home/mrtg/public_html/performance> ~/tmp/avgperf.pl AUE
LAX
total daily data files read: 931
AUE to LAX latency (min/avg/max): 157.603/166.720/999
mpetach@netops:/home/mrtg/public_html/performance>


That is predicated on having good direct links, which is
eye-wateringly expensive if you're used to US data costs (data going
from
China to Australia via San Jose...  aaargh).  Then again, hosting within
Australia is similarly expensive, so splitting your presence isn't
going to
help you any from a cost PoV.

It's not really a matter of eye-wateringly expensive, so much as simple
basic existence; there's no direct Sydney to southern Asia fiber, at the
moment; the best you can do is hop through Papua New Guinea to
Guam, and then back across into southern Asia.  (or overshoot up to
Japan, and  then bounce your way back down from there).

Matt




Current thread: