Database Reference
In-Depth Information
Clients
DirectFLOW
1
2
3
1
Request to get & store data files
File request matched to map of object
locations across StorageBlades
Transfer data directly
2
3
DirectorBlade
Cluster
Single Virtual
Namespace
StorageBlade Cluster
Figure 2.4 Panasas storage uses the DirectFlow installable file system to
enable clients to consult clustered metadata servers for maps to the location
of data on storage that can be directly and independently accessed in parallel.
storage and clients. Because this object storage protocol is not yet distributed
with every release of Linux, Panasas offers a file system client module, called
DirectFlow, that can be installed onto each client in the compute cluster. All
of Panasas storage then shows up in the namespace of each DirectFlow client
at /panfs. Accessing a file in /panfs for the first time causes DirectFlow to
communicate with the metadata managers to request permission to access
the file, the reply to which will contain a map of the object locations of the
file on storage servers. With a map, and the security capabilities that come
with it, the client can directly access object data in parallel and rarely needs
help from metadata managers, which could not support nearly as high a total
bandwidth if all of the data had to pass through the metadata managers as it
does in traditional distributed file systems like NFS or CIFS. Since, however,
not all machines at a site are in the Linux cluster, the metadata managers
in Panasas also implement NFS and CIFS servers to enable other machines
to get to the same data. Because Panasas clients, including the metadata's
NFS and CIFS servers, have consistent client caches and use a distributed
locking protocol to maintain correctness during concurrent access, all NFS
and CIFS servers offer the same exported data, a property called clustered
network-attached storage (NAS).
Search WWH ::




Custom Search