Hacker News Reader

Comment by joshhart | Hacker News Reader

$136M is a lot of money for a distributed file system. Can anyone compare GlusterFS to HDFS? I think they're booth GoogleFS clones but I'm not sure on the details except that HDFS uses a name node and is written in Java.

View on HN

> Can anyone compare GlusterFS to HDFS? I think they're booth GoogleFS clones but I'm not sure on the details except that HDFS uses a name node and is written in Java.

I don't think GlusterFS is GoogleFS clone. As far as I know HDFS isn't meant to be a mountable file system. GlusterFS is a near POSIX-compliant FS that can be mounted with FUSE. The FUSE mounts are then exported via NFS or CIFS. At least that is how it was 2 years ago when I tried it. I think NFS is now a native export option of GlusterFS volumes via the NFS Translator. In my mind GlusterFS is more like Isilon than HDFS.

I don't really like the fact that GlusterFS is in user-space. I think Ceph is a more promising project: http://ceph.newdream.net/

Having GlusterFS in user space gave us a lot of advantages (development cycle time, bugs not panicking the kernel etc) ... besides in a distributed scale out architecture doing things in kernel or user space does not make much of a difference for performance as either network or the disks become the bottleneck.

As you mentioned the storage volume can be mounted via NFS too.

They're not really directly comparable. GlusterFS implements standard POSIX semantics and is directly mountable on a client. On the backend it can use another file system (ext, XFS, etc) where it actually stores the data. It's highly configurable in a variety of different server distribution schemes similar to RAID levels you'd use on local disk.

GLusterFS 3.3 even supports the HDFS API.

http://www.gluster.com/community/documentation/index.php/3.3...

$136M is a really good deal for a distributed file system that actually works and is used by many thousands of companies.

Reply on news.ycombinator.com