Hi Jens,
My suggestion is to go with column table with that many rows. Remember - row tables need to completely fit into memory don't have any compression! You also saw the consequence of going too far 
. You also have to account for any indexes that you might have created for your memory consumption.
HANA is fairly efficient at inserts and updates with column tables as long as you make sure you are doing the delta merge on a regular basis. If you run into memory pressure, HANA will page data back out to disk from memory. With HANA One on AWS, you can also optimize the log processing speed by creating a provisioned IOPS drive (solid state) and then mounting it to your HANA One instance. You can then configure HANA to use the new drive as the log drive. This is essentially how the HANA appliances are configured and the way SAP originally envisioned configuring HANA One. The problem is that a provisioned IOPS drive is much more expensive than a standard EBS Raid0 configuration and that SAP didn't want people to have sticker shock when they got their bill at the end of the month from AWS. But, that's the great thing about AWS - if you need the horse power - they can generally deliver.
Regards,
Bill