Before we start configuring samba, however, we need to give some thought to how the shared disk space is going to be used. Even used machines these days commonly have 4 Gb . drives, which minimizes the necessity for network storage to supplement local file storage. As a rule, network file storage should probably best be thought of as enabling file sharing and as representing a repository for individual file storage for backup and accessibility from other locations on the network. This is, of course, nothing new. Fille sharing is one of the first network applications to gain wide-spread use in the personal computer world, and peer-to-peer file sharing on local area networks has been common since the introduction of Windows 95. While adding such capability would rarely be considered the prime rationale for building the mosix cluster, if there is any need at all for file-sharing on the local network the benefits of centralizing the sharing into the cluster justify the action.
Samba is, of course, a tool that can be configured at various levels of sophistication, depending on the complexity of the desired end. This discusssion will begin with a fairly simple implementation, to illuminate the primary considerations involved, and build on that discussion to include the use of utilities and more sophisticated installations. As with most things, there is no need to build a solution more sophisticated than your requirements, because it will likely require a higher level of maintenance. At the same time, constructing an overly simplified solution will constrain the applicability of the overall network. Frequently, determining the appropriate level is an iterative endeavor, if only because as users acquire more sophistication they will help define higher-level applications that will drive the need for higer levels of sophistication in the network services provided. Working through the installation in this fashion will help you to develop the tools required to make this leap.
A NOTE:Although I'm not yet into the benchmarking phase, I've started to do a little playing around with putting a load on ralphzilla with large compilation jobs and then starting two large copying operation to the shared directory on ralphzilla-raider. It's fairly clear that under this kind of operation saturation of the hub does become a problem. The congestion can manifest itself from the standpoint of the user as difficulty connecting to the share and potentially interrupted copying operations. I rather think that moving from a hub to a switch, even a relatively low-end switch, will substantially mitigate these symptoms. In the interim, just recognize that the potential exists for this kind of difficulty under heavy load with this configuration.