News for hackers. Stuff that matters.
Re:Comparing Filesystem Performance in Virtual Machines by notacoward on Monday March 02, 2015 @08:21AM (#9131685)
One point that might be hidden in the OP is that hypervisors are often configured to lie about whether data has been committed to disk. This is certainly true of most public clouds. As a distributed file system developer I run a lot of I/O tests in the cloud, and I've gotten many results that are impossible to explain any other way. People I know who work at some of these providers have practically admitted it. In a cut-throat industry, any provider who actually did the right thing here would get hammered on performance and price (because they'd be unable to pack as many instances onto each host as their competitors do). It's something to be aware of when you run any data-intensive application in a public cloud, or when you're configuring hosts in a private one.
Re:Comparing Filesystem Performance in Virtual Machines by falcolas on Monday March 02, 2015 @09:09AM (#9131885)
I don't know if it's possible to nail down a vendor for comment, but are we able to back this up? It would be good to know for cases like running an ACID database.