It's interesting, but the granularity is kinda weird. >300ms is a nonstarter for all kinds of web apps here in Japan.
I worked on a web app project here where we absolutely could not use EC2 APAC (when it was only Singapore), and ping times to EC2 were 60ms. We had to put physical boxes in a Tokyo data center. It was only when Amazon opened their Tokyo region (8ms ping times from real users) that the customer was okay with the speeds.
So is would have liked to see finer gradations of these results. Still cool to have the info, though.
Agreed, < 125ms is typically an OK range (for most things) but < 300ms is really nebulous. There are apps I would stop using fairly quickly if 300ms was the norm.
Thanks for the tip on Tokyo data center for EC2, I've seen a lot of complaints for the Singapore data center in general with terrible times into China and Tokyo before and wasn't sure how the people actually in those regions felt about the response times.
That's interesting. So U.S.-East seems the best option for supporting both the U.S. And most of Europe. Although if you're slightly less concerned about the U.S. Then the EU seems a reasonable conpromise.
One thing I doubt is the uniform ping times within countries. For example, do we really think ping times are uniform across the breadth of Russia?
0-300 is a pretty wide category, though; 300ms is not a great response time. Better would be to target 100ms. A logarithmic scale would probably have served better in the color-coded maps.
If you have a global market, and only want to operate in one region (a practical necessity for many people, given the technical challenges of doing otherwise), then US East is almost always the best option.
Wow, wish I could have seen that talk. I recently got similar results by doing client-side load balancing: 27% improvement, on average, for users that were not close to EC2 East.
Based on this data, you can have really fast responses around the world if you provide your services in all regions.
However...
It is really difficult to get their cloud services to work across Regions (AMIs not shared, security not shared, RDS, Cloud-formation only works in one region, etc.). If you want to run in multiple regions, you need to use EIBs for your database servers (mysql, mongo, etc.), setup your own replication and fail over, etc.
All in all, though, a great service (though at first, I didn't like the entire degraded server problems but realized it has forced the issue of not assuming servers will not fail).
I would point out that global replication is one of the things that makes CouchDB's master-master replication so tasty.
You can't accomplish the same thing with Mongo in a R/W environment since you can only write to the master in a replica set. You could have multiple read-only slaves in other geographical locations, but that isn't always helpful.
You can also somewhat fake it with a much more complex mongo-replica-set-per-region sharding configuration, but then you don't have the same data available in each region in each replica set.
PostgreSQL in 9.1 added master/master streaming replication which is nice. Not sure what MySQL does in these situations.
This is one of the biggest failings of SimpleDB that I think keeps it held back from being a huge success, the region dependency of it and the seeming lack of support from Amazon (the only product whose forums frequently go a week without a reply).
Had SimpleDB has an optional global redundancy for quick read/writes in all locations, I think it would have been the killer product a lot of people were looking for.
Instead I have to roll my own with a fully connected CouchDB graph of servers. Ahh well.
That doesn't look right to me. My country (South Africa) gets ping times of <250ms for pretty much anything international, yet the chart shows 500-750ms. Where is the other 250-300ms coming from?
Enlightening, thanks. A shame that none of the big names have a data center in South America. I've a project I'd like to bring to Brazil at some point and 300-500ms is not very good. What will it take to open up a node in Sao Paulo or Buenos Aires? Is it lack of traffic or taxes holding up progress?
I find it odd that GAE has lower average response times in Canada than it does in the USA. However, as others have said, the granularity of the study leaves a lot to be desired.
Why is it that Canada is consistently getting better or equal performance than the US? I would think that most traffic requests from Canada are routed through the US.
I worked on a web app project here where we absolutely could not use EC2 APAC (when it was only Singapore), and ping times to EC2 were 60ms. We had to put physical boxes in a Tokyo data center. It was only when Amazon opened their Tokyo region (8ms ping times from real users) that the customer was okay with the speeds.
So is would have liked to see finer gradations of these results. Still cool to have the info, though.