GFS2

I notice this 2009 post so late. It is about Google resorts to GFS2 (known as Colossus inside Google) for two reasons: (1) a single-master design makes it inconvenient to failover; and (2) the queue, into which thousands of operations are dumped, can optimize for throughput but is not great for latency. Of course, this post has nothing about how GFS addresses these problems.