As you note, there are a number of costs associated with EC2, beyond the cost of running the instance, however, the 'instance-storage' mentioned in your question is not one of these.
Each EC2 instance (except the t1.micros) comes with ephemeral storage, that is included at no extra cost (neither for I/O nor storage). However, this storage does not persist - as soon as the instance is stopped or terminated, the contents of the ephemeral storage is erased. It is only useful for transient purposes (e.g. temporary files, some types of caches, intermediate files in build processes, etc)
You will, therefore, require an EBS volume to store your files - both for core software and your web application. I usually favour using at least 2 EBS volumes - one for the root volume, which I expect minimal changes to, and one for everything else (databases, logs, mail, application + user data, etc).
I can't judge your storage needs, but software (operating system + LAMP stack + mail, etc) will likely be under 2GB. I presume your application is fairly small - which means that the unknown is user data (databases, files, etc) - I'd say 10GB is likely enough to start with in most cases.
- EBS volume #1 - root volume - 4GB (allows for growth, some extra space on the drive)
- EBS volume #2: 10GB
- If you use a t1.micro, add EBS volume #3 - swap space - 1GB
Total: 15GB = $1.50/mo ($0.10/GB/mo)
EBS Estimated IOPS:
- Root EBS - 2-5 IOPS
- Data EBS - 10-30 IOPS
- Swap EBS - < 1 IOPS
Total: 13 - 36 IOPS - cost: $3.42 - $9.46/mo (~$0.2628/mo/IOPS)
50GB/mo (your estimate) = $0.120*49 = $5.88/mo (1st GB is free)
An elastic IP is free as long as it is attached to a running instance
Backups - EBS snapshots
I'd estimate around the same used space as your total data stored (the data is compressed and each snapshot is differential, but they do add up over time), so presuming you start with only 8GB of data + 2 GB on your root drive = 10GB (obviously, swap wouldn't be snapshotted);
10GB = $1.25/mo ($0.125/GB/mo)
Total, excluding EC2 instance costs:
Elastic IP - $0.00
EBS storage - $1.50
EBS I/O - $5.00 (a number in the middle of the range)
EBS snapshots - $1.25
Bandwidth-Out - $5.88
Now, you need the actual instance costs. If you will be running the machine continually (required for a web server) - I'd suggest a reserved instance. (730 hrs/mo, on average)
- On demand = $0.02/hr = $14.6/mo
- 1 year term (heavy utilization): $62 + 0.005/hr = $8.82/mo ((0.005*24*365+62)/12)
- 3 year term (heavy utilization): $100 + 0.005/hr = $6.43/mo
- On demand: 0.085/hr = $62.05/mo
- 1 year term (heavy utilization): $276.25 + 0.02/hr = $37.62/mo
- 3 year term (heavy utilization): $425 + 0.02/hr = $26.41/mo
Depending on the choices you make, your costs come out to somewhere between $20/mo (3 year reserved t1.micro) and $80/mo (on-demand m1.small).
5 requests per second is a fairly small amount - even a t1.micro should be able to handle that. I would suggest trying your application on that instance type first (AWS does offer new customers one year of free t1.micro usage) - if you need a larger instance type, it is a simple matter of stopping (not terminating) your instance, and modifying the instance attributes, you can be back online in under 5 minutes running your new instance.
I am not a Java person - but if you can use a lightweight web server (nginx, lighttpd) instead of Apache you will get much better performance out of an EC2 instance (the primary problem with Apache is its memory usage, especially on a t1.micro).
If you use the t1.micro, I would advise against using a 64-bit operating system, it has a limited amount of memory, and the 64 bit registers considerably increase the needed memory with no tangible gain.
Also, I if you do not have a particular preference for an operating system, I'd suggest Amazon's Linux AMI - it is based on RHEL/CentOS with all the unnecessary parts removed, and is officially supported (and they keep their repositories up to date!).
Don't bother with Amazon RDS - most people find its performance to not be that great, and it doesn't offer much by way of 'automatic' functionality (e.g. you still have to manually setup replication, etc, and the normal limitations apply).
A word of caution regarding the t1.micros - their performance is highly variable - if you can keep your CPU usage low, you will be fine, but if you find it spiking, the performance will be awful. The other instances have much more stable performance.
Finally, while you want a good idea of how much it will cost, don't over-estimate. It is easy to speculate that you will need a mammoth of a server to run your application, but if you find it coming to that, you can always find a different provider later on, or shift to a different instance type. With the cloud model, you don't need to factor in future scale as much, because it is fairly easy to scale.
You can try the AWS Cost Calculator to play with the numbers (although, personally, I prefer doing it by hand).