Amazon Sydney EC2 Latency

[Mikel Lindsaar]

Some people here might be interested in this:

How to shave 150ms off content served from EC2
http://reinteractive.net/posts/25-how-to-shave-150ms-off-content-served-from-ec2

Mikel

[Jason Kotchoff]

That's really cool Mikel.

Does anyone know whether there might be any heroku plans to be able to choose a data center for a heroku app?

Also, whilst on the regional EC2 thread.. For a rails web app with a global audience running on EC2, are there any opinions out there on how to architect regionally specific application servers to optimise user response times?

ie. would regional servers (eg. an australian, a european, an asian mongrel instance etc.) even achieve time savings for server-side rails code if they need to talk to a central once-off database (eg. in the USA)..  Or would this make sense with master-slave database servers..

If anyone has come across a good write up on distributed rails architecture for the cloud, that would be really interesting!

[Jak Charlton]

Use something like Akami to front your app and globally distribute

And Heroku repeatedly say no plans for anything but US East ... annoyingly

--
You received this message because you are subscribed to the Google Groups "Ruby or Rails Oceania" group.

To view this discussion on the web visit https://groups.google.com/d/msg/rails-oceania/-/Kg_rosIdvqEJ.

To post to this group, send email to rails-...@googlegroups.com.
To unsubscribe from this group, send email to rails-oceani...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/rails-oceania?hl=en.

[Pat Allan]

Heroku were asking at least some users questions about a potential data centre choice, with a focus on Singapore or Japan. Obviously, neither is ideal for those of us in Australia (which I certainly made clear when I was filling out their survey).

Beyond that, I'm using Amazon's CloudFront more and more often to at least ensure assets are served from somewhere close.

--
Pat

[John Dalton]

On Wed, Nov 14, 2012 at 4:15 PM, Jason Kotchoff
<cornflake...@hotmail.com> wrote:
>
> Does anyone know whether there might be any heroku plans to be able to
> choose a data center for a heroku app?

Can't say anything for Heroku, but Engine Yard will have Sydney as an
option very soon.

> ie. would regional servers (eg. an australian, a european, an asian mongrel
> instance etc.) even achieve time savings for server-side rails code if they
> need to talk to a central once-off database (eg. in the USA).. Or would
> this make sense with master-slave database servers..

Having an app hosted in Sydney talking to a DB server in the USA will
be much slower than having both servers in the USA - this is because
your app will typically make multiple DB calls per request, so you're
multiplying the delay rather than reducing it.

That said, with smart caching and local DB read slave[s] a read-heavy
app might work out with this approach. Your writes back to a central
DB will still be slow, and your local read slave will be much more
susceptible to replication lag - however for some apps this isn't much
of a problem. It really depends on the app, and is something you'd
have to test!

--
http://johndalton.info
http://www.linkedin.com/in/johnrdalton

[James Healy]

On 14 November 2012 09:52, John Dalton <john....@gmail.com> wrote:
> Having an app hosted in Sydney talking to a DB server in the USA will
> be much slower than having both servers in the USA - this is because
> your app will typically make multiple DB calls per request, so you're
> multiplying the delay rather than reducing it.

An alternative would be to host a nginx/varnish/something proxy on EC2
in Australia that maintains a keep alive HTTP connection to the
upstream app servers in the US.

Users will have low latency to the local caching server, and despite
all requests still tromboning to the US and back, the response time
will be faster because the user avoids the multi-RTT TCP handshake
across the pacific.

I guess the trick is deciding whether the speed improvements are worth
the extra complexity in the architecture.

James

[Daniel]

+1 for having frontside cache servers in multiple regions.

As long as you have reasonably cacheable content it can be a pretty substantial speed improvement.

[Simon Russell]

> An alternative would be to host a nginx/varnish/something proxy on EC2

> in Australia that maintains a keep alive HTTP connection to the

> upstream app servers in the US.

This reminded me of:

https://www.cloudflare.com/railgun

Which I'd been meaning to find out more about; anyone tried it?  Seems like an interesting idea.

 

[Jak Charlton]

Apparently it costs around $800 per month per website

--

You received this message because you are subscribed to the Google Groups "Ruby or Rails Oceania" group.

[Simon Russell]

I think it's $200, but yes, it's not something you're going to use everywhere...

[Bruce Wang]

On Wed, Nov 14, 2012 at 8:02 PM, James Healy <ja...@yob.id.au> wrote:

An alternative would be to host a nginx/varnish/something proxy on EC2
in Australia that maintains a keep alive HTTP connection to the
upstream app servers in the US.

This sounds like what CloudFront does, and Sydney already has a edge location.

see http://aws.typepad.com/aws/2012/05/amazon-cloudfront-support-for-dynamic-content.html

Cheers,

Bruce

--
simple is good
http://brucewang.net
http://twitter.com/number5

[Andrew Stone]

I love Heroku... it _really is_ awesome! but my customers are local....

Next week I will move my stuff over to EC2 here in Sydney... sure its a step backwards for platform as a service... But honestly the latency from east coast US sucks.
The local EC2 DC is very quick... some corporates in the Sydney CBD were getting sub 5 millisecond pings on Tuesday.

Stonie.

[Steve H]

On Friday, November 16, 2012 2:06:03 PM UTC+13, Stonie wrote:

Next week I will move my stuff over to EC2 here in Sydney... sure its a step backwards for platform as a service... But honestly the latency from east coast US sucks. 
The local EC2 DC is very quick... some corporates in the Sydney CBD were getting sub 5 millisecond pings on Tuesday.

It's good enough for me to seriously consider hosting NZ websites too - ~27ms from Auckland on business fibre.

Although it was mostly unreachable for me from until yesterday, so I'll see how reliable it is first.

[Bayan Khalili]

Ninefold were here with similar services well before Amazon.

I can't really see much of a difference between them in terms of their offerings. Are there any?

Regards,

Bayan

On 16 November 2012 18:28, Krasimir Angelov <krasimir...@gmail.com> wrote:

 sure its a step backwards for platform as a service.

Have you tried EY? 

To view this discussion on the web visit https://groups.google.com/d/msg/rails-oceania/-/BhbnN_01vFMJ.