I'm a big AWS fan in general... but RDS, not so much.

@RolandoMySQLDBA has pointed out are some pretty good points against it.

The only advantage I see in RDS compared to MySQL on EC2 are the ability to do point and click snapshots, clones, and point-in-time recovery, but these are not nearly sufficient to make up for the loss of control and flexibility and they most certainly don't justify the price being higher. RDS is sexy in some ways, but you can't ultimately trust what you can't ultimately fix, because you can't get to all the moving parts.

I don't like not having the SUPER privilege. I don't like not being able to tail the error log. I don't like not being able to run "top" or "iostat" on my database server to see how the cores and drives are enjoying the load. I don't like not having access to the federated storage engine. I don't like the thought of paying for a hot standyby (multi-AZ) backup master machine that I can't even leverage as a read replica. Sure, there are perfectly reasonable explanations why all of these constraints have to be in place for MySQL to be successfully packaged and sold as RDBMS-in-a-box. The only thing is, RDBMS-in-a-box "solves" a whole series of problems I don't have... and gets in my way, causing new problems.

But the absolute deal-breaker for me with RDS is the complete lack of access to the binary logs and replication. Binlogs, especially row-based, are a fantastic recovery tool for minor disasters, but they are of no help to you if you can't access them. Want to configure an on-premise server at your office as a read-replica of your production database in RDS? Something to take local backups from, do reporting, have on hand for disaster recovery should something unthinkable happen to your data that lives in RDS? That's an idea that is simultaneously obvious and brilliant.

Oops, sorry, direct access to replication is not available. Sure, you can pay for read replicas... but only as other RDS instances. Not a value proposition in my book.

Update: One Significant Change in RDS for MySQL 5.6

I still prefer running my own server (even in EC2) as opposed to running RDS for a number of reasons, including the lack of support for User-Defined Functions, the inability to use the Federated Storage Engine, and the inability to have the one extra connection available for emergency access... however...

Amazon has made a significant change in MySQL 5.6 for RDS that eliminates one of my major objections -- perhaps my largest objection: the binary logs are now accessible and you can run a non-RDS instance as a slave, or connect other utilities to the server that read the binlog stream.

http://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/MySQL.Procedural.Exporting.NonRDSRepl.html

Officially, the documentation indicates that they are exposing this so that you can set up a slave for the purpose of doing a live migration -- you synchronize the foreign future master server from the existing RDS instance using mysqldump, then connect it to RDS as a slave to get a live feed of updates via the replication stream until your application is migrated to the new master -- but unofficially, you can do this on an ongoing basis as long as you don't expect them to support you... which, to me, seems reasonable.

This was confirmed in a recent Webinar, in a conversation that begins at around 56:45:

"You can keep it in a replicated state indefinitely...

"...as long as you take the responsibility to maintain the replication..."

"We are not preventing you from doing ongoing replication if that's what you want."

This new capability was enough for me to drop my blanket objection to using RDS in our public-facing website-backing MySQL instances, where we don't use FEDERATED or some of the other things as much or at all.

So I'm still not in favor of it, but I'm no longer against it, since having a live stream of the binary logs puts me ultimately back in control of the data in real time and the responsibility for ensuring that no transactions are lost in a catastrophic outage is back with me, because I, as the DBA, am back in control -- which is exactly how I want it. Having a third party vendor to point fingers at, or file a lawsuit against, or whatever, doesn't get your lost data back if it disappears down a black hole inside a black box.

Management seems to like the "idea" of RDS and doesn't object to the cost difference, so we are now launching all new web sites with RDS behind them.

The point and click point-in-time recovery, I admit, is a nice feature in RDS... it doesn't alter or disrupt your existing machine -- instead, it fires up an entirely new instance, using the backup that was closest-in-time to the selected point in time, and then applies the necessary binlogs to bring that new machine forward to the point in time that you've specified.

Related to this, but in the other direction, it's also possible, now, to use a similar strategy to migrate a live MySQL database into RDS... you can connect an RDS master (presumably, typically, this would be a newly-deployed instance) as a slave of an existing system so that the RDS instance has the live version of the data at the time you migrate into it. Unlike access to the RDS binlogs for outward replication, which only works in 5.6, the inward replication is supported in RDS beginning with 5.5.33 and 5.6.13.

It's the file size that's the problem.

Essentially there's a timeout expiring somewhere, the server logs will tell you where.

When the timeout expires, no more data is produced, and after a while with no data the connection gets closed as per normal.

The last time I had to tackle something like this (Which is a long time ago now), I wrote a small program that basically pulled the data out of my server in chunks using various SQL criteria.

This is absolutely NOT the best way of doing it, but when you have a table that's so big and takes so long that things start to time out, it's one of the only ways I found to make it work.

The other approach you have is to use an ETL tool such as Pentaho and do a server to server transfer, again using a custom query to do it a chunk at a time.

If you MUST do it using PSQL, then try to break it up. Instead of one massive 150gb file, try doing the xfer a table at a time or similar.

Also, think about local resources. You might get better traction if for example, you can dump the file to a server where you have file system access that's inside amazons network, such as an AWS instance. If you have a local resource, set up an AT job, dump the DB to that instance (optionally G-zipping it in the process) then transfer that file to your PC using a web browser of FTP connection.

Tackling the initial problem head on, your not going to know which approach you have to take until you can figure out why it's timing out.