One thing that helped us a lot was setting

checkpoint_completion_target = 0.9 # up from 0.5 default

So the checkpoints get written out more evenly and cause less and shorter "stop the world" events. This was especially painful on Amazon EBS (which is a relatively bad idea by itself btw) with large shared memory size. Decreasing this actually improved performance for us.

I had exactly the same article in mind when reading the OP one.

Exactly the opposite message. Who to believe?

They are not incompatible messages at all. This article says that SSDs delete stuff fast; the other says you can't reliably delete particular things. Basically, they're unpredictable. You can't reliably expect a particular file to be purged, and investigators can't expect data to stick around. So it might stick around, it might not, and no one has direct control.

I'd say every programmer should read 1984 by George Orwell, to always keep in mind the big picture implications of their work.

When I first read it, I got shivers: "this book is about google!"

And after Amazon deleted the very same book from every Kindle remotely ... you could not ask for a more ironic example of how well Orwell predicted things.

> When I first read it, I got shivers: "this book is about google!"

If you read it again today, you would think it's about Apple.

We are building a similar site with a bit different approach, ours is a free site which encourages people to create their events at their homes or public places, also bring together communities and create a framework for small businesses, a bit like farmers markets or the like. Check us out at: http://eatwithme.net/