Most places likely aren't using Oracle at Amazon's scale, either.

I suspect that one of MySQL, Postgres or Microsoft SQL Server can fill in for Oracle if you have access to the source code and can make changes yourself. (If you've bought software that is Oracle-only, then it's not an option -- but it wasn't an option for Amazon, either). I suspect that in a lot of cases, this will pay for itself, but it might take 10 years to do so.

Depending on the size of your install, there are companies that are very good at converting over something like Oracle to Postgres for probably much less than you'd expect. With the price of oracle licensing, your break-even could be in a year or two.

Wouldn't surprise me if a large fraction of those installs could be replaced with sqlite..

>Microsoft SQL Server

Out of the frying pan, into the fire.

MSSQL's really not too bad. Yes, it's a commercial offering, but there's nothing inherently evil about charging money for software, and you do get what you pay for.

I've worked in both MSSQL shops and PostgreSQL shops, and, while I wasn't in charge of the budget, I wouldn't be at all surprised to find out that, once you factor paying your programmers and DBAs for their time into the equation, the MSSQL shop ended up having lower overall costs associated with its choice of DBMS.

Have you worked on MSSQL licensing in the age of Virtualization and per-core licensing? It's gotten VERY complex according to some of my friends that still run it.

It seems they have changed their pricing strategy:

    Oracle: What is your IT budget? We'll take it. (all of it)

    Microsoft: Hey, at least we're not as bad as Oracle.

MS licensing is somewhat expensive, but easy to understand and relatively benign. Oracle licensing chokes you like the Drakh keepers from Babylon 5.

I'm currently working on a mssql shop, and on aws it's more than twice the price vs postgres or aurora. I don't see how you can justify the price difference

I'm not sure I follow this logic. Wouldn't both databases require the same amount of care and feeding? Are you suggesting that you need less DBA's per MSSQL database than for a PostgreSQL or Maria database?

Yep. Because the tooling is stronger. The developer ergonomics are better. The performance is better, meaning you'll be able to do more with fewer servers. MSSQL has a boatload of useful features that PostgreSQL lacks. (Clustered indexes and table-valued parameters are the ones I like to fixate on.) And so on. I can only personally speak to the developer story, but my experience has been that the productivity difference is huge. There's a night and day difference in the information you can get out of an EXPLAIN on each of them, for example, and I would get detailed emails warning me if a MSSQL query plan unexpectedly changed in production. With PostgreSQL you're more likely to need to build your own tooling for that sort of stuff, or just wait for a user to complain. On the DBA side, I'm hazier on the details, but I've seen PostgreSQL DBA's sinking hours' or sometimes even days' worth of effort into grinding through tasks tasks that an MSSQL DBA can bang out by spending a couple minutes clicking through a wizard dialog.

If you're a tiny startup, the differences won't amount to all that much, and it will be hard to justify MSSQL's price tag, which is fine, because, yeah, you can't afford it, anyway. If you're an enterprise or otherwise have large database needs, the differences can very quickly become apparent, and easily worth the money.

I have worked with both databases over a decade. The biggest difference between SQL Server and Postgresql is neither tooling or performance(both are excellent), but the community and culture.

By far, PostgreSQL has more knowledge sharing and documented edge cases. Not to mention that the official PostgreSQL documentation is probably one of the best documented software systems ever. This makes PostgreSQL a lot easier to learn and use and that also affects the amount of money you need to spend.

SQL Server has some annoying edge cases, licensing is a real pain point these days, but so are the technical issues like lack of schema support in the JDBC driver. Other things are that you cannot pipe the backup. I would love to take backup of an Azure SQL service to my local machine with SSH.

Instead of a table-valued parameter, wouldn’t you use an array of a composite type in PostgreSQL? What are the relative good and bad points?

Yeah. And you don’t need to create a custom type to handle it as a parameter. So it’s easier.

The only thing SQLServer has over PostgreSQL is multiply resultsets. That’s it. I really wish that existed in PostgreSQL to prevent round trips.

That said, if you restrict yourself to the PostgreSQL subset of Aurora, you will be able to move to any Postgres implementation down the line.

Yes, they charge you a fee to move your data off. However, that is not the limiting factor for people moving off of Oracle. (And if it is, they're probably not looking at AWS in the first place.)

I would protest the point that you have to be as large as Amazon to produce your own database-as-a-service solution, especially if you are only building it for yourself. Throwing some automated tooling around Postgres and any arbitrary linux VM hypervisor API should get you a lot of the same value.

It doesn't take a ton of magic to allow a developer to spin up a pg cluster with a few clicks via some custom web tool. Reporting and operations are trickier to handle, but if you don't need to satisfy the same customer base as Amazon does (i.e. you are your only customer), you could probably pull this off at nearly any scale.

That said, I will certainly grant you the cloud vendor lock-in angle. It truly is one of the biggest ploys of this decade. My organization takes great pride in only using the most fundamental of cloud resources (compute instances, block storage, DNS, VPN, routing). If we had to evacuate AWS to on-prem or some other provider, it would not be a huge ordeal. For those who wound up with their entire business stuck firmly inside vendor-specific things like Lambdas, you have my deepest sympathies.

They don't really charge you a fee. The only thing you are charged is data transfer out cost of whatever amount of data you transfer out. And that price is the same whether you are transferring data out for backup, for use by your internal systems, etc. Or for leaving to another vendor.

Also, as far as I know, all cloud providers charge for data transfer out.

There are plenty of alternatives that don't lock you to Amazon. One of my teams just migrated from oracle to MariaDB and saw big performance gains. We happen to be using AWS for hosting, but it's Maria, we can go anywhere.

Thank you for mentioning the departure fee... That just took AWS off the table for me unless I'm pushed into it by other (paying) parties.

It's not really a fee. It's just the price they charge you for data transfer to external systems. This is known as Data Egress charges and all cloud providers I know off has it. The cost of egress will depend on which service you are using. Some services have lower data transfer out costs, some have higher, etc..

There is no departure fee. As others have pointed out, this is simply data transfer.

I believe the parent is referring to the price of the bandwidth that would be required for transferring data from AWS to somewhere else.

can you give an eli5 of postgresql's offering vs oracle db? Why isn't there an open source equivalent right now? What 'extra stuff' did Amazon have to build to make it 'more' than the open source stuff available right now.