News for hackers. Stuff that matters.
Re:New Dyno Networking Model by cachvico on Thursday May 02, 2013 @02:44PM (#5646384)
Funnily enough this happened with Heroku and Facebook just a couple of days ago - a bunch of apps got blocked from using the Graph API until Facebook removed the banned AWS IPs.

https://addons.heroku.com/proximo does what's needed, but at a serious price for high volume.

Re:New Dyno Networking Model by sehrope on Thursday May 02, 2013 @03:37PM (#5646664)
Wow that is pretty steep pricing. The lowest tier that seems usable is the $75/mo one. Then again if you're locked into Heroku then even $1,250/mo isn't that much for piece of mind and added security. I'm sure whoever is paying for it is happy it's available.

This was one of the reasons we handled server deployments ourselves for our startup. The cloud version of our app runs in production on AWS in a VPC so all outbound traffic is NATed through a single public IP. With reserved instances it costs about $22/mo for the NAT gateway and setup was pretty straightforward.

Re:New Dyno Networking Model by agwa on Thursday May 02, 2013 @03:50PM (#5646745)
> all outbound traffic is NATed through a single public IP. With reserved instances it costs about $22/mo for the NAT gateway and setup was pretty straightforward.

I've considered doing this myself so I'm curious: are you getting good network performance through the NAT gateway? What instance type are you using?

Re:New Dyno Networking Model by sehrope on Thursday May 02, 2013 @04:00PM (#5646841)
No issues to speak of so far. Our app databases are in the VPC itself so our own traffic does not get routed through the NAT gateway but I haven't heard any issues from users either.

At the moment we're using an m1.small for the NAT itself though if you have perf issues you can bump that up as well. I would guess a c1.medium would be more appropriate though as I said we haven't had any issues so haven't considered changing anything yet.

Here's a speed test through from a m1.small through the NAT (27.5 MB/s):

    $ curl -o /dev/null http://speedtest.wdc01.softlayer.com/downloads/test500.zip
      % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                     Dload  Upload   Total   Spent    Left  Speed
    100  500M  100  500M    0     0  27.5M      0  0:00:18  0:00:18 --:--:-- 29.2M
Here's a speed test from a m1.small vanilla EC2 instance outside the NAT (38.2 MB/s):

    $ curl -o /dev/null http://speedtest.wdc01.softlayer.com/downloads/test500.zip
      % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                     Dload  Upload   Total   Spent    Left  Speed
    100  500M  100  500M    0     0  38.2M      0  0:00:13  0:00:13 --:--:-- 34.1M
Re:New Dyno Networking Model by agwa on Thursday May 02, 2013 @04:12PM (#5646910)
This is great info - thanks! :-)
Re:New Dyno Networking Model by rdl on Thursday May 02, 2013 @04:02PM (#5646858)
I can't believe all people don't use VPC whenever they have >1 instance in AWS.
Re:New Dyno Networking Model by ceejayoz on Thursday May 02, 2013 @04:21PM (#5646967)
It's actually now the default for new accounts.

http://aws.typepad.com/aws/2013/03/amazon-ec2-update-virtual...

Re:New Dyno Networking Model by jorde on Thursday May 02, 2013 @03:23PM (#5646593)
This was also totally random and gave us some serious headache: one dyno at the time starts failing with Graph API. Restart helps but only for some time until Heroku randomly cycles your instances. Would be happy to pay little extra for a reserved IP.