The development.
Development of a number of important issues addressed in the design of
distributed file systems. They relate to the functionality, and performance of
semantics. Various file systems can be compared with each other, understanding
how they address these issues : The namespace - Some of distributed file
systems provide a single namespace such that each customer uses the same track
name to access that file. Other systems allow customers to create their
namespace is shared by mounting podderevev arbitrary directories in the
hierarchy of files. Both methods have their attractiveness. operation for the
conservation and preservation of free-Preserving the status Server stores
information about transactions between customer demands and uses this
information on to the correct service subsequent requests. Such requests as open
or seek linked to climate conditions, as someone must remember information about
what files are opened by the customer, as well as the removal of all open files.
In states without pay each request is "self-sufficient" and the server does not
support the steady-state customer. For example, instead of supporting the offset
of the open file server can require the client to the removal of each read or
write. Servers with the conservation status are faster because they can use the
knowledge of the client to significantly reduce network traffic. But they must
have a range of mechanisms for maintaining a coherent state of recovery from its
refusal. Servers without pay of more simple in design and implementation, but
did not give a higher yield. Semantics division-Distributed file system must
determine the amplitude, which is used when multiple clients simultaneously
turned to one file. Semantics UNIX requires that all changes made by one client,
would be visible to other customers, when they pose a systematic challenge to
read or write. Some file systems provide a "semantics session (session semantics),
in which changes are made available to other customers on the basis
granulirovannosti system calls open and close. Some systems are even weaker
guarantees, for example, the length of time that must pass before the changes
will fall to other customers. Methods dial-in a simple client-server model using
remote operation, in which every action initiated by the client and the server
is a simple agent, which serves client applications. In many distributed systems,
especially systems that maintain state, the server plays a much more active role.
It not only serves the needs of our customers, but also participates in the
mechanism to ensure coherence, informing clients of all cases where it cached
data are unreliable.
Network file system.
NFS Sun Microsystems NFS submitted in 1985 as a means of transparent access to
remote file systems. In addition to publishing the protocol Sun has licensed its
basic implementation, which was used by various suppliers to NFS porting to
different operating systems. Since then, NFS was actually industrial standard,
which is supported by all the options really UNIX, as well as some other systems
such as VMS, and MS-DOS. Architecture NFS is based on client-server model. File
is a machine, which exports a collection of files. Clients are machines that
have access to these files. One machine can for different file systems to serve
both as a server and a client. However, the NFS code is divided into two parts,
which allows only client or server system. Clients and servers interact with the
remote procedure call (rpc-remote procedure call), which operates both
synchronous requests. When the annex to the customer tries to remote file, the
kernel sends a request to the server, while the client is blocked until a
response. The server waits incoming, processing them and sends back responses to
customers.
|