SSHFS Multiplexer (SSHFS-MUX) was born because we live in the cloud era, where you are likely to have multiple/many machines and you want to share data among them.

Originated from, but Beyond SSHFS

SSHFS Multiplexer (SSHFS-MUX) originates from SSHFS, but is more than SSHFS.

For example, by SSHFS, you can:

$ ssh host:/target/directory /local/mountpoint

    directory          mountpoint
     /     \     ==>    /      \
  data1   data2      data1    data2

where via mountpoint, you can manipulate remote files as local ones.

By SSHFS-MUX, you also can:

$ sshfsm host1:directory1 host2:directory2 /local/mountpoint

    directory1     directory2         mountpoint
     /          +         \     ==>    /      \
  data1                  data2      data1    data2

where directory1 and directory2 are properly merged in to one global namespace at /local/mountpoint.

This can be also achieved by using SSHFS and UnionFS/UnionFS-FUSE. However, using SSHFS-MUX is more straightforward, easier, and efficient.

For more details, please refer to Usage.

Optimal Transfer Rate in Wide-Area Networks

Using SSH to transfer large data in long-fat network does not fully utilize the network capacity. This is because the transfer buffer is fixed in OpenSSH implementation, which is illustrated here and can be solved by HPN-SSH.

If you feel tedious to patch OpenSSH and you trust your network, then you can try using raw socket instead of SSH. Following chart shows how the optimal transfer rate is achieved by using raw TCP link.

Point-to-point transfering 256MB data in different block sizes over the long-fat link.

Building Your Own Scalable, High-Performance Distributed File System On-the-Fly

Using SSHFS-MUX is more than that. You can even build a scalable and high-performance distributed file system on arbitrary machines in a few minutes. GMount has been proved practically useful by both micro-benchmark and real-world data-intensive applications in multi-clusters environments.

Following chart shows that mounting a global file system takes less than 10 seconds on over 300 nodes across 12 sites.

Following charts shows that GMount has scalable metadata and I/O performance in WAN (up to 32 nodes across 4 sites).

GMount is more complex than SSHFS-MUX, but still very easy to use comparing to other conventional distributed file systems. Here we only present its usage, the detailed design and evaluation can be found in following publications:

• Nan Dun, Kenjiro Taura, Akinori Yonezawa. GMount: An Ad Hoc and Locality-Aware Distributed File System by Using SSH and FUSE. In Proceedings of The 9th IEEE/ACM International Symposium on Cluster Computing and the Grid (CCGrid '09). (pdf)