|
The Pulsar file system team has so far implemented three server applications: a web server, an NFS server, and a server for an experimental user-level distributed file system. We have run Pulsar on clusters of up to eight nodes, and we plan to start running sixteen-node experiments soon. Initial performance is encouraging; Pulsar's lower layers can write data at up to 31 MBytes/s on four distributed disks. With more tuning we expect to push near our hardware's SCSI-bus limit of 40 MBytes/s. The current system survives node crashes without metadata corruption; we are beginning to implement support for redundant storage and node-failure availability. We have concrete plans to make Pulsar the main storage system for the Princeton CS Department. |
