1

As a student, I run a cluster using ubuntu server in my college. As you may already know, each computer of this cluster, have HDDs, and I'd like to create one single HDD for the cluster...

I'm talking about something like a virtual HDD, that have all the hdd clusters as a single one. The problem is, I can't find much on the internet... and I don't know how to search for it exactly (the right words for a good search).

Could someone help me?

s3lph
  • 14,314
  • 11
  • 59
  • 82

1 Answers1

1

Network filesystems

I assume it's about a compute cluster consisting of consumer PC hardware - there is no load balancing or high availability involved

In general terms, what you are looking for is a network filesystem.

Technically, a network filesystem consists of a daemon on the server which has the real harddisk, and software on the client machines that provides something like a "virtual harddisk" on the client;
The software makes it look like there is a local harddisk, but accessing it will be redirected to the harddisk on the server.

Implementations

There are many implementations of network filesystems:

nfs

A common implementation used under Linux is nfs

It is well suited for "general" uses in the default configuration; For your use case, it may be relevant that it provides lots of options to tune it for specific needs, because a cluster with a given load may have very unusual harddisk access patterns.

See How to set an NFS network? and Setting up NFS HowTo.

sshfs

An alternative to nfs could be sshfs. It provides a filesystem on the client using the normal sshd, which allows remote logins by ssh and file operations by scp.

Compared to nfs, it may be slower in general, but on the other hand it's much easier to manage, and the server side sshd, is usually available by default.

SMB/CIFS

Another alternative is SMB, commonly used with Windows, implemented by samba.

Cluster specific filesystems

Because the network filesystem used in a cluster can have a big influence on the performance, the network filesystem is an important part in the design of high performance clusters. To maximize performance for specific classes cluster aplications, various specialized cluster filesystems exist. These can also handle data in other structures than filesystem trees, or provide unusual file locking services.

Volker Siegel
  • 13,065
  • 5
  • 49
  • 65