32 file servers to serve ~480TBs of disk. Seriously? 4 file servers, GPFS and 4 MD3060e. Thats about a petabyte of usable storage with a streaming throughput of about 40 gigabits a second.
"modern" clustered filesystems are mainly just toys. if you want speed, use lustre, if you want sexy software defines awesomeness with next level raid, use GPFS.
ceph is horrifically slow, and only has a rudimentary posix interface.
gluster is just well, terrible.
first things first a SOHO office cannot support a clustered filesystem, unless one of the people happens to be a storage specialist.
yea I hear lots of noise about how self healing they both are. That is mostly fancy talk for "I don't have backups"
Supposedly they both do HA. But yeah, its not something I'd want to support.
If you look at gitlab's setup, and their proposed setup (https://about.gitlab.com/2016/12/11/proposed-server-purchase...)
32 file servers to serve ~480TBs of disk. Seriously? 4 file servers, GPFS and 4 MD3060e. Thats about a petabyte of usable storage with a streaming throughput of about 40 gigabits a second.
"modern" clustered filesystems are mainly just toys. if you want speed, use lustre, if you want sexy software defines awesomeness with next level raid, use GPFS.