Document Home

PC's

Ralphzilla is currently comprised of five machines, four of which are in the p75-p100 range and one of which is a p200. Conceptually, it looks like this, five machines connected to a hub, one of which, the cluster controller, is dual-homed, with one connection to the wider network and one to the internal network representing the mosix cluster.


Following is a short summary of the machines involved.

Ralphzilla pentium-90 32 Mb. RAM 1.2 Gb. & 512Mb hard drives cluster controller
Ralphzilla-raider perntium 200 32 Mb. RAM, 2 Gb IDE drive, 2 Gb. SCSI drive, 4 Gb SCSI drive
Ralphzilla-faxa pentium 100, 32 Mb. RAM, two 850 Mb IDE drives, two 412 Mb IDE drives. 28.8 external faxmodem
Ralphzilla-free pentium 100, 32 Mb. RAM, 1 Gb. IDE drive, 850 Mb. IDE drive
Ralphzilla-b-free pentium 75, 32 Mb. RAM, 1.2 Gb. IDE drive

As you can see, none of these are very powerful by today's standards. The disk subsystem of each of the machines, with the exception of Ralphzilla-raider, was configured to balance the capacity of the disk system with system performance without using any larger drive than necessary. In the context of this agency, an older, slow, 2 Gb. IDE drive is still a resource to be used in workstations, if possible. Therefore, the machines were configured with the goal of supporting the applications required without unnecessarily constaining the machines. A remote boot configuration was not chosen because the default configuration of the cluster involves a relatively slow network interface, and placing traffic on that connection that was not required was not regarded as adviseable. Neither was relying on a slower drive than necessary, for obvious reasons.


The two exeptions to this rule are Ralphzilla-raider and Ralphzilla-faxa. Ralphzilla-raider was configured to support the highest level of disk IO possible. (I know that sounds incongruous in this context ... maybe you should think to yourself "get the most out of what we have".) This machine is, in effect, the cluster data-center, supporting both the shared disk space and the postgresql database server. (Ralphzilla-raider is, in fact, a stand-in machine taking the place of a similar, but somewhat more strongly configured, machine that is currently playing a role analogous to the one played by Ralphzilla-raider in this configuration. Since that machine is in daily use, I didn't want to interrupt service to use it directly as the configuration is developed and tested. As the system configuration is finalized, that machine will be folded into the cluster and the machine now Ralphzilla-raider pulled out.) Configuring the machine that plays this role within the cluster in this manner is especially important in that one goal is to achieve the highest sustainable level of throughput for the system as a whole as utilization levels climb.


Ethernet Hub

The hub used to connect the machines is a David Expressnet 10 Mbps hub, probably current ca. 1994. As will be discussed in more detail, this is anticipated to be a limiting factor in the possible configurations of the cluster, less because of the limited speed of the hub than because as shared ethernet media in a cluster involving a great deal of communication between machines, it is probable that at some point a threshold will be reached that will result in saturation of the cluster network from collision-induced traffic. In fact, that's part of my rationale for choosing this hub as a starting point. Evaluation of these limitations should identify the point at which the performance requirements of the cluster would make upgrading this element of the cluster important, and what tradeoffs could be made in cluster software components to balance components for which a certain level of performance is desired against those for which performance limitations would not represent a problem in any given context. Ultimately, I hope to make that element of this discussion the most meaty.



Modems

The modems used to provide fax services are both external, a USR Courier 28.8K and a USR Courier 33.6K. Modems can very tricky with linux, if for no other reason than that many, if not most, of the internal modems sold today include only some of the requisite functionality, relying on the operating system to provide the rest. While a knowledgeable user might be able to get a winmodem configured for use with linux, it has been my experience that external modems, by their nature required to incorporate all modem functionality within their shell, represent a much more reliable choice. All other factors being equal, an external modem will also outperform internal winmodems. It is, however, important to recognize that the maximum speed for fax transmission is 14.4 kbps, a speed easily achieveable by four to five year-old modems. As 28.8K and 33.6K external modems are readily available for approximately $5USD on Ebay, and would likely be newer than 14.4K modems, their purchase is justifiable.


On the other hand, it does not make sense to purchase a 56K modem for this role. The $30-$35USD a reasonably-priced external modem would cost could purchase four or five 28.8K modems, leaving you with several replacement modems. If your interest involves setting up a box for shared internet access, you should be following a different path. The design of this cluster attempts to maximize the useful life of legacy hardware by providing a scalable platform for a set of common share network uses. The kinds of demands experienced in the use of a shared modem or pool of modems could also be satisfied by legacy hardware, but the requirements of that system would be in substantial conflict with the requirements of a system providing fax services. If such a system is desired, it should be implemented separately. For most of the country, the price of just two phone lines puts one well within proximity to the price of a DSL or cable internet connection, a far more efficacious manner of providing a shared internet connection to multiple users. Should those services not be available in a given location, the best choice would be to employ a technique called modem-doubling, should your ISP support it. Again, use of this architecture to provide shared internet access is both not within the scope of this discussion and an inappropriate use of this architecture in the first place. Should that be your interest, I would suggest starting here .