AWS - How much do you need?

In looking through the posts, it seems several people have FreePBX running in Amazon Web Services - But I would like some specific information from people running there such as:

  1. What Instance type are you using? From what I do on Hyper-V, it seems a t2.medium instance would be enough for most of my customers (<50 phones) - What instance are you using and why, and how is it working for you?

I base this assumption on this: https://www.thewebmachine.net/ec2-deployment-guide

  1. What does your monthly cost work out to be? We have just finished setting up all the redundancy and fail-over at our CoLo, but we are up to a total cost per month of about $800 - we don’t have to sell that many hosted systems to pay for that, but is that high in relation to AWS? Low?

  2. What has been the reliability of AWS for Voice?

Any real info would be great!

Thanks!

Greg

Do not use AWS if you are looking to provide hosted FreePBX services or really any type of voice services.

  1. AWS is costly.
  2. AWS oversells.
  3. AWS immediately puts you behind NAT. Your end users will already be behind NAT, as their service provider being being NAT yourself just introduces a world of problems that you shouldn’t really have. Not to mention how it can impact your peering with upstreams.

It just is the wrong choice for a VoIP service provider.

Thanks - We are not, but it is tempting - always trying to keep my options open…

@BlazeStudios Sorry I don’t mean to be rude but I need to ask, have you even used AWS before ?

  1. More costly than what ? What are you comparing it too.
    I’m running multiple phone systems , each with 50-100 users, each on instances that cost me roughly $40 per month each. (t2.medium) - which I’d hardly call ‘costly’

  2. No they do not oversell. This ‘line’ comes from people who don’t understand how their pricing structure works and then intricacy with burstable instances. A few of my good friends work for AWS in their data centers and if anything they severely undersell. I can assure you there is no lack of capacity.

  3. This is absolutely incorrect. In fact for you to be behind a NAT you would need to go out of your way to set up a separate NAT instance in your VPC and attach it to one of your machines. Unless you are very determined to turn it on, there is no NATing of any kind that occurs by default on a instance. Once you attach a elastic IP, it’s straight 1:1

2 Likes

Interesting…May have to revisit this.

Not sure why anyone would want to jump through hoops to “try” to get Freepbx to work in AWS or anything else for that matter. Freepbx has worked with CyberLynk for a number of years now and they definitely has Freepbx hosting down. I have used them for years and can say I would never recommend anyone else for your primary hosted server.

2 Likes

CyberLynk does not peer with Comcast (at least they don’t list them in either of their datacenters) and that is a serious issue for someone wanting to provide hosted PBX because of Comcast’s deep penetration in the United States. What this means is that if you have a Comcast subscriber with a network problem then there’s always going to be an intermediate network that your provider and their provider will point the finger at and you will get nowhere with.
AWS by contrast as the largest cloud provider in the world, can usually be used to “immunize” yourself from blame. So when the subscriber with a crappy line disconnects from your service in disgust and goes spewing about how much you suck on the social media, you can easily respond with “we use the largest hoster in the world, Amazon” And you can also use that leverage to tell the subscriber “your ISP sucks, it’s not us” if the subscriber wants to fix it instead of just spewing.
Unfortunately it is pretty amazing what some end users out there will do in an effort to save money on Internet connectivity. It wasn’t too long ago for example that Sprint pulled the plug on CLEAR and the screams on the forums were terrific from subscribers who were using their crappy Internet service to save $5 a month and claiming it was just as good as a land line.
I’m sure CyberLynk is fine for small businesses and corps who have an IT person who can manage that relationship but when you are selling to end users, many of who will fight you tooth and nail defending their gear even when it’s obvious that their gear is the problem, I can see why someone would want to use AWS.
Big Brand names do matter in these situations, I know it sucks but that’s the way it is.

AWS works fine.

The only downside is getting FreePBX into AWS. Something I wish Sangoma would take care of by offering a supported AMI. There is an AMI but its not free.

To get started with AWS you have to do the following:

  1. Build a VM from the FreePBX distro using something like VirtualBox.
  2. Export the VM from VirtualBox.
  3. Upload the Exported VM to an S3 bucket.
  4. Use AWS CLI tools to import the VM.
  5. Deploy the import VM (AMI)

Like Jszaszvari said, you are looking at about $40 per VM. A safe estimate is between $40-60 depending on your storage requirements.

Performance / audio is rock solid.

Never had any issues with NAT

It’s worth looking at but might not be for everyone.

Slight side note. I use FreePBX on AWS and have a few paid Sangoma modules activated. I recently added a 2nd hard drive to the server and messed up the fstab and the server wouldn’t boot. Not a huge problem because I had an AWS snapshot of the server so I just restore it. Only issue I had was the FreePBX/Sangoma activation detected it as new hardware and I had to reset the hardware lock so I could reactivate it using the same deployment ID. You only get a few hardware lock resets so could become a problem!

1 Like

@tmittelstaedt I have been using CyberLynk for years now. 75% of all of our clients have Comcast and their network, although one of the largest as you had quoted, is also one of the worst. Run PingPlotter from a non-Comcast internet connection and monitor points on the Internet that are connected to Comcast and monitor points on the Internet that are not. I do this myself from many locations and the amount of Jitter and Latency is daunting.

There is no problem with CyberLynk hosting when a customer has Comcast, If there is a problem on the Comcast network it does not matter if CyberLynk has a Comcast connection or not.

In the area I’m in Comcast and Centurylink are the 2 mammoths. However you cannot get CL copper other than PPPoE and I have trouble with raccoon/IPsec VPN routers on it likely due to the lower MTU. OpenVPN is fine, though, although it also encapsulates. As a result of this corporate lan2lan VPNs built with IPsec on Comcast show higher throughput than lan2lan VPNs built with openvpn on a PPPoE circuit all other things being equal. (probably due to fragmentation)
I don’t know if Jitter is really an issue with VoIP today (unless it’s really really bad) since newer phones seem to have adequate jitter buffers, but I’d be interested in what you see as daunting latency. I’m on Comcast myself and even behind 2 routers plugged into it, with a bunch of servers contending with it, I don’t see any higher latency than around 25ms and most of the time it’s around 10-15ms. I can say from experience though that Comcast has issued a lot of crap cable modem gear in the past and you are always best off using your own cable modem and own router with them. And, signal levels are critical and not all the Comcast install techs really understand that having too high a signal is as bad as too low.

AWS FreePBX from TheWebMachine Networks is the only supported solution you are going to get on AWS. We are the only certified Sangoma Partner offering a “true-to-distro” option on AWS. Sangoma has expressed on multiple occasions that they have NO interest in building their own public cloud images. This is where we come in. We charge for AWS FreePBX in order to cover the extra development costs in providing a truly stable version of FreePBX that conforms to all the same standards as an ISO installed distro build while ensuring proper optimization for the AWS platform. We also provide free personalized support above and beyond the forums with that software charge. Make no mistake, Sangoma also takes their fair cut of what we charge on AWS.

But more to the point, FreePBX has been working on AWS flawlessly, in one form or another, for years now. Their NAT is of no concern, their network is of no concern, and we have systems running on AWS with 500+ extensions and hundreds of queues with no issues. We also released our own clustering support that leverages other AWS services (RDS/EFS) to create one of the most resilient disaster recovery solutions available for FreePBX. Sure, AWS may appear to cost a little more on the surface, but you can’t knock their infrastructure. They are the #1 public cloud despite their pricing for a reason. Besides, once you leverage things like Annual Software Subscriptions and Reserved Hardware Instances, the costs become extremely competitive with the rest, while still having access to the endless array of services they provide.

As far as jitter is concerned, it’s almost counterproductive to enable any jitter buffers in today’s VoIP environment. Given the wide variety of codecs available, with many softphones (especially those designed for the mobile market) being able to automatically determine which codec to use based on bandwidth availability, it should be fairly easy to find a codec that provides enough balance between quality and bandwidth requirements to not need the extra latency introduced by jitter buffers. In all honesty, adding a jitter buffer to an already shoddy connection (which probably already has some latency issues) will just add to the “walkie talkie” effect to devastating ends. For those with absolutely terrible connections or very limited speeds, the GSM codec is rather handy. Sure, it’s old-school cellular quality, but it’s better than “Well, I was thinkin…oh, go ahead…ok, yeah, I…sorry, can you…” all the time.

I have system with 50-100 extensions on $5 Vultr instances. with zero issues. $1/month more gets nightly backups of the instance.

For the OP wanting to provide hosted phone services and needing some kind of geo redundancy, it is not a solution, but otherwise DO/Vultr class VPS solutions are one of the best ways to run a PBX on a public “cloud” service.


A little off topic: how do you connect the phones to the PBX? VPN or a Dynamic DNS address?

Many businesses that have 50-100 handsets have a static IP, but if not, yes Dynamic DNS.

1 Like

I asked because I did not see a Static IP for this instance, and I was wondering how you connect the phones…

Oh the PBX side? Yeah it has a fixed IP. That would be begging for trouble. That said, I still always use a FQDN for everything.

1 Like