In the cloud, Mumbai is a long way from Asia Sending packets from Singapore to Mumbai over AWS? Fetch a coffee, the latency is horrible – according to cloud performance data released yesterday. Metrics outfit ThousandEyes ran tests from network probes deployed into 27 data centres globally, collecting connection data from 55 AWS, Azure, and Google cloud regions. The …
AWS architects its network to push traffic away from it as soon as possible. A possible explanation for this behaviour is the high likelihood that the AWS backbone is the same as that used for amazon.com
Probably true. Alternative explanation is that 'hot potato' routing has been the standard BGP traffic management policy pretty much since BGP peered out of a cloud. Problem is hot potato (aka closest exit) may mean dumping traffic at a congested local port. For a transit service provider who wants to sell on quality, that traffic policy may not be the best idea. Especially not with the big 3's capacity buying power vs other ISPs or telcos.
Asia's also always been an interesting traffic engineering challenge. For a long time peering was via JPIX or HKIX which leads to Indian, Singapore or other large parts of Asia taking a long route via peering. Efforts to grow that in places like Mumbai or Singapore are growing, but customers need to look closely at how their ISP(s) peer with the clouds in any region.
AWS architects its network to push traffic away from it as soon as possible.
All of the cloud providers tell us not to worry about capacity, etc. - whatever is required, the cloud will scale up accordingly in order to provide.
I've long suspected that each provider has finite capacity and their strategy for handling over-demand is to procure services from another provider.....and that in times of extreme demand, that second provider brings in capacity from a third provider and so on.
How much load is required before the final overflow provider needs more capacity and the whole thing loops back on itself?
"AWS's Singapore region is, at most, 30 milliseconds away from a user probe in America, Australia or Sao Paulo, but on average 50ms from probes in Asia and more than 60ms from Europe."
Unless someone's changed the laws of physics, no, it's not. That graph shows latency variation, not actual latency. Singapore -> West Coast CONUS is something more like 150-180ms.
I wondered the same, but they have measured latency variation rather than latency.
Having dealt with a number of sites in India relying on connectivity to Singapore, there are gotchas around whether you can get traffic onto the SMW links. Hit those and every ones happy - if maintenance is happening on one of the links, you quickly find out if a provider has saved money by not buying committed bandwidth...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
