3

Is there such a thing? I don't care about performance, I just need to be able to have access to the same block device from two computers. The block device does not have anything fancy (like reservations), so it's pure read/write device without caching or other nasty tricks.

George Shuklin
  • 296
  • 2
  • 11
  • Are you sure you want a shared device rather than a network (NFS, SMB) or parallel file system (e.g. glusterFS)? A shared device is a SPOF. – symcbean Jun 28 '23 at 10:12
  • Of course, a single shared device is a SPOF. I would better look at something like Starwinds VSAN. It can turn underlying storage into the shared one. Also, it can simply make high availability based on a few servers. https://www.starwindsoftware.com/vsan – Stuka Jul 08 '23 at 15:51
  • Yes, I do. I need block device sharing, not a redundancy. – George Shuklin Jul 09 '23 at 06:38

3 Answers3

4

Global File System v2 (GFS2) from Red Hat is probably the most popular clustered file system for Linux. See:

https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html-single/global_file_system_2/index#ch-overview-GFS2

There's Oracle Cluster File System v2 (OCFS2) and IBM General Parallel File System (GPFS) as well. See:

https://blogs.oracle.com/cloud-infrastructure/post/a-simple-guide-to-oracle-cluster-file-system-ocfs2-using-iscsi-on-oracle-cloud-infrastructure

https://www.ibm.com/docs/en/gpfs/4.1.0.4?topic=guide-introducing-general-parallel-file-system

Red Hat can't scale beyond 16 nodes due a fact it has no dedicated metadata hosts and every host has his very own copy of all the meta. Making long story short: Performance actually decreases as you have to keep everything in sync, distributed lock traffic becomes monstrous (no recent public benchmarks published, sorry about that!). Oracle's file system is designed to handle Real Application Cluster (RAC) so works well with a big files, small file performance isn't that great. IBM's file system isn't 100% open source so I wouldn't deploy it in the open source environment.

Everything else is quite exotic.

BaronSamedi1958
  • 13,676
  • 1
  • 21
  • 53
0

They are named cluster filesystem https://en.m.wikipedia.org/wiki/Clustered_file_system . There are different products with different features. For instance: GFS2, OCFS2, Lustre

NoNoNo
  • 1,963
  • 14
  • 20
0

Caching is a layer above an actual problem, conflicting writes to the disk. Say node A and node B are presented the same disk. They try to write different things to the same blocks at the same time. Very likely this corrupts the file system.

Instead, use a network file system like NFS or SMB, which support multiple node access. Many full featured storage systems can present volumes as file shares.

Cluster file systems such as GFS, GPFS, or OCFS, also exist to address concurrency problems. Their designs tend to require nodes have a distributed lock manager and fencing mechanisms (those fancy SCSI reservations) to make multiple node block access safe. Historically, cluster file systems have the reputation of being more difficult to implement, compared to a file share with logical file access going through one server.

John Mahowald
  • 32,050
  • 2
  • 19
  • 34