Thinking about dumping AWS for local server

For a testing server, AWS seems too expensive. A one-year “reserved” convertible 16 GB server is $1,626.00 (t3.xlarge). For that money, you can get a cheap PC and Windows Server 2019. From there, you just set up DDNS to update your registrar’s IP address (assuming you have dynamic IP like from, say, Comcast).

The disadvantages of having your own “PC” is that you have all that (potentially malicious) network traffic coming into your router. Sure, you would direct it to just the Windows Server computer, but it’s still concerning to me.

I’ve tried both these approaches now and am still not sure which is best overall given each pluses and minuses. Both are compelling and both give me pause.

I know folks on the forum that would argue for both positions. If you get a Comcast business setup, for example, that would possibly be more expensive than AWS after your initial two year lock in rate.

Comments? Suggestions?


1 Like

Can you get a second dsl or cable connection? You could completely isolate the test machine from the rest of your network.

AWS services are cheap if you only pay for them when they are in use. For database interactions that’s not possible. You need the DB to be highly available.

The benefits are the security and network availability. If you add those prices into your costing the differences come down.

Other considerations are the flexibility. Do you need a 16GB instance or can you use a smaller instance with S3 storage. Do you need t3.large at all times? If your have identifiable periods of peak usage and low usage you may be able to utilise a lower cost instance for off-peak.

Depending on your business, those options may not make sense, or the cost of implementing them may make them unattractive. People spending a lot of money on computing, hardware and network resources can really save money.

I used FileMaker Cloud on the AWS service. For me, I prefer to not have to also be managing, maintaining and be responsible for a server. I use a t2.medium with 4GB ram, but our needs are relatively small, at least for now.

You’re right. I really just needed a testing server.

I set up a new AWS t3.large which is only 8 GB (Windows 2016 after Windows 2019 would not work). I like the fact that should I need additional resources I can just change the instance type. One “gotcha” I never understood (until now…) is that every time you start your instance Amazon nicks you an hour. So, during all the set up, I should have just left the instance running instead of thinking I could stop it and start it repeatedly to save money.

Thanks Malcolm. :slight_smile:

1 Like

That’s an AMI on the Amazon Marketplace, right? What are the differences are between FMS cloud and a server you set up yourself. For me, I need to setup database servers, install things like Tomcat and micro-services so I really need control over the whole machine.

FMCloud is a virtual machine based on linux with a linux version of FMS. However, Cloud is not feature identical with the standard edition of FMS. You’d have to check the docs to see which pieces are missing. In v1 I recall that it was CWP and scheduling but I’m not up-to-date.

There’s nothing stopping you from customising your Cloud instance. It is plain old linux. However, any update or rebuild from a clean image is going to wipe your work. You’d need to have a good build command that is able to load your custom work after any sort of system change.

Since I had read many issues with it on the FMP forum, I was just mildly curious about the cloud solution. I already just re-setup an AWS FMS solution that works fine for my testing needs. Personally, I would not use the cloud solution or use an FMP hosting company.


I setup the FileMaker Cloud instance in AWS following the procedure laid out (very nicely) by Richard Carlton (YouTube). The video is a bit old, but it wasn’t when I first set this up. :wink: It was VERY helpful. Like I’ve said, I’m on a part-time FileMaker developer and not much more than ‘Beginner’ status. My only redeeming quality is that I’ve been using FileMaker at this level since FileMaker 3 or so.

Hope that helps. :slight_smile:

For what its worth, I have been really happy with the FileMaker Cloud solution. It has allowed a single part-time beginner level FileMaker developer to quickly and inexpensively build and host solutions for a company of about 20 users with hardly any issues or hiccups (that were not really just my development :wink: ).

Just curious…what are you paying for hosting?

Do you host 24/7 or do you turn off the instanced sometimes?

Do you have a reserved instance?

My t3.large instance (Windows 2016 Server), if I ran it 24/7, assuming 30 day month would cost:

                        ( 0.11/hr.

× 24
× 30 )

= $ 79.2/ mo.

Setting up FMS is as simple as installing a, well, an installer.

Not suggesting the cloud approach isn’t a good one…I’m just curious what your costs are.

I did a very unscientific rank order survey of a number of top FileMaker developers, about their preferences for which deployment model they prefer. Not a single one, opted for FMCloud/AWS, even though most of them have that as at least one within their supported stable of deployments.

The overwhelming (>90%) of the respondents opted for a Virtual Private Server as the #1 option, with a dedicated hosted computer as #2 @ ~75%. FMCloud came in dead last. That does not mean there are not appropriate use cases where it makes sense, but it was definitely not preferred.

Ran a benchmark test about 8 months ago, using 2 remote servers in 2 different locations, logged into a remote desktop at a 3rd location.

Server 1 was a quad i7 gen2 machine with 16gb of RAM and a SATA III 250gb SSD.
It was connected to a 100mBit up / 100mBit down, nearly unlimited bandwidth (lots of pipes), synchronous network connection.

Server 2 was a quad i7 Gen 7 machine with 32gb of RAM, and an NVMe/PCIe 512gb SSD (6 times the disk I/O speed of server 1),
It was connected to a 300 mBit down / 35 mBit up, asynchronous cable connection.

Installed the same FM database on both servers, and logging in from a remote machine (as to have minimal impact to the network I was on concurrently with Server2).

Server 1 - the older, slower machine, ran FM remotely, at about 3x the speed of the newer, far faster server. Nothing else was installed on either server, outside of FMS.

The (ad-hoc, unscientific benchmark) moral: in WAN connections, network speed is paramount. There are design methods to mitigate the WAN performance costs, but given a typical FMP solution running over a WAN, network speed and available bandwidth (two different parameters) is critical. And synchronous connections are seldom available outside of a data center.


Thanks for the test results. Interesting to see that. FMP is really a two-way conversation. And it’s important for customers to know that when they are choosing providers and plans.

Interesting. I found exactly the opposite. AWS’ ease of use, configuration, and cost (compared to say, $300/month for Comcast “Business”) is what everyone I know uses.

I also use AWS’ “Elastic Beanstalk” for applications that only get charged when used.

In any case, a VPS would be preferred for me as well. Whenever FMI creates an install for a Linux machine (say CentOS, which is what my VPS uses), I’d jump on it too as that wouldn’t cost me any additional hosting. I’d still use AWS for Elastic Beanstalk, Lambda, Hosting Websites like Drupal and WordPress (for $5/mo), their S3 where I store Terabytes of data, etc.

For my use, FMS needs to be accessible on the Internet not just some local office for a group of users. Adding the Internet access opens up all these cans of worms.

For my testing needs my t3.large AWS instance costs me about $8/month.

Thanks for your reply.

Love benchmarks as they’re so objective.

I did a benchmark recently also to help a client understand how much faster JDBC was than even FileMaker itself.

Here are the results:

Test 1:

(Local) FileMaker, script only copy 10,000 sales records from one table to another using local SSD drive (on 2012 MacBook Pro, 16 GB, SSD):

36 seconds.

Test 2:

Using free FMI JDBC drive with same FMP database:

7 seconds.

PCT Difference: 81%

Test 3:

Using JDBC with same FMP database but on AWS:

11 seconds.

In another test, using PSOS only gave me a one second improvement not the dramatic difference JDBC gave.

1 Like

What is going on there? That is remarkable.

1 Like

JDBC is just tons faster in most DB operations than FMP itself (not talking about native Finds, of course). That’s it. I’ve done tons of bench-marking examples like this and posted many on the “other” forum.

Some things that help using JDBC code:

  1. Avoids slow FMP scripts
  2. Avoids extremely slow FMP script loops
  3. Enjoys Java’s Pre-compiled PreparedStatements
  4. Enjoys Compiled Java code

It all adds up to subtract lots of time. :slight_smile:

(Using the same JDBC code with MySQL was 2X faster (or 1/2 the times above) still than JDBC with FMP.)

JDBC code is cross-platform and would work with any other database vendor assuming the same schema with almost no modifications (mainly login syntax but not SQL).

I’ll have to set up a few experiments to get comfortable with the best way to use this.


it depends on how You want to use the server. Is it more or less locally (from Your office) or from anywhere in the world, accessed by many users, etc.

I’m not really familiar with windows and I set up an AWS system with FMS twice for testing
If You can manage it to access AWS for some relatively short time, it is quite cheep - otherwise it becomes expensive
Futhermore, I failed with moving the AWS from ‘west coast’ to Frankfurt. There was no human-help available…

I played with the costs - and finally got a Mac Mini. I also checked Windows systems (there are several models available with about the ‘Mac Mini’ form factor). Since one needs a server os for running fms, the Windows system becomes expensiver than a Mac System - and the setup is way more difficult (for me), means more costy

Today, I’m running some multi-file solutions on that Mini-server and it is really smooth - but we are just 2 or 3 users at the same time.

Only port 5003 was open at the beginning. At the moment, there is no more port-forwarding, the Mini is not available from outside. Since my machine at the office is running all the time, I installed ‘Real VNC’ on that machine and I get access from everywhere, no matter if iPad, iPhone, whatever
If more than just one or two people are using FMS, this method is no choice, but here, I’m happy with that - for 2 years now

1 Like