Distributed Filesystems: Fraunhofer vs. Gluster (2014)
moo.nac.uci.edu
Distributed Filesystems: Fraunhofer vs. Gluster (2014)
1–9 of 9 posts
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#2> "We tried Ceph about a year ago and it wasn’t nearly ready for production use"
Given that it's a year-old article, that 2 year-old impression is probably very outdated. I would have loved to see Ceph benchmarked there as well.
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#3There is also little known but very cool MooseFS http://www.moosefs.org/
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#4> "We tried Ceph about a year ago and it wasn’t nearly ready for production use" Given that it's a year-old article, that 2 year-old impression is probably very outdated. I would have loved to see Ceph benchmarked there as well.
The article is really outdated.
There is no "FhGFS" anymore.
http://www.beegfs.com/content/
What i don't like are metadata servers. I mean mostly I would start out small, with every project. But the "real" cluster file systems don't follow this tradition. The bare minimum of a solid production environment is probably more than 5 nodes, which is really huge for using it as a layer for content distribution in the small scale.
I mean its cheaper to rent s3 / the new aws nfs solution or any other s3 like storage solution, than running a server with a distributed file system by yourself and I'm just considering the running costs, no maintance etc...
Its really hard to do this on premise on a really small scale and just for availability. Most people will mostly just buy this stuff from san/nas vendors if they could afford it, if they have a system that needs to have this built in your pretty much fucked with ugly hacks / solutions.
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#5Not sure if they've worked on it, but 3 years ago Gluster performance during rebuild was about 1/100th of in normal operations making it effectively impossible to do a rebuild of a production cluster if you lost a node.
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#6There is also little known but very cool MooseFS http://www.moosefs.org/
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#7> "Glfs has a severe problem with small writes, both with writing small files and small writes into large files. FhGFS does not have this bottleneck."
GlusterFS has implemented many smallfile performance enhancements and with the latest release I am seeing a 20-50% improvement on smallfile performance. There are more improvements in the pipeline:
http://www.gluster.org/community/documentation/index.php/Fea...
I would be curious to see these tests run on a properly tuned(readahead on bricks is important for spinning disks and read workloads, I didn't see any mention of brick tuning in the article) and on the latest version of glusterfs with the smallfile performance enhancements.
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#8Also Arvados Keep (http://arvados.org) is a content-addressed distributed file system.
Re: Distributed Filesystems: Fraunhofer vs. Gluster (2014)
#9> "We tried Ceph about a year ago and it wasn’t nearly ready for production use" Given that it's a year-old article, that 2 year-old impression is probably very outdated. I would have loved to see Ceph benchmarked there as well.
The article is really outdated. There is no "FhGFS" anymore. http://www.beegfs.com/content/ What i don't like are metadata servers. I mean mostly I would start out small, with every project. But the "real" cluster file systems don't follow this tradition. The bare minimum of a solid production environment is probably more than 5 nodes, which is really huge for using it as a layer for content distribution in the small…
> I mean its cheaper to rent s3 / the new aws nfs solution or any other s3 like storage solution
Sure if you never pay for bandwidth.