| From: | Tzeentch tzeentch666@*********.net |
|---|---|
| Subject: | Virtual Realities 3.0: Clustered Systems (Beta Rules) |
| Date: | Mon, 24 Jul 2000 18:59:31 -0700 |
Edition - Virtual Realities 3.0. Even minus TM material it should weigh in
at about 200+ pages of Matrix goodness.
**Note that after reading more on this subject the below rules are in DIRE
need of updating but I might as well show it to people for feedback. These
rules "plug into" my own hardware construction and programming rules. As I
said, this was written before I read too much about the subject so obviously
its kinda whacked out.
***For OS Rating read "MPCP Rating" for the time being.
****Core Rating is a concept of hardware capability that will be explained
when I revise my programming and hardware rules.
CLUSTERED SYSTEMS
A number of mainframes can be linked together with a high-speed dataport to
effectively network their operations and function as a single, much more
powerful system.
[Single-system image]
LIMITATIONS
Operating System
Each system that is to be connected in a cluster must have a
clustering-capable mainframe OS. All of the OS ratings must be equal.
[Cluster-aware Mainframe OS's are at +2 to their Multiplier]
System Links
Every system in the cluster must have enough System Links to connect to each
of the other systems. If the systems are physically separated then each
system link must connect to the other systems link over a connection capable
of handling the maximum system I/O (see System I/O, p. xx) of the highest
rated system in the cluster.
For example, four systems are clustered together. The highest-rated system
has an Mainframe OS rated 10. Thus its system I/O speed is 400 MePS. All
four systems must have at least that much free bandwidth on the lines
connecting them.
[This part will probably go away, or take the role of RL "cluster
acceleration" techniques.]
Location
If physically separate, all parts of the cluster must reside in the same
RTG. If on a PLTG all of the clustered systems must be located on the same
PLTG.
[Given signal delays and the insane problems this would cause for cache
coherency and memory access this will probably be revised to "in the same
LTG"]
Disconnects
If any machine in the cluster is taken offline without a proper shutdown,
then the clustered host crashes, losing all data in active memory and
fragmenting all information in storage memory. Each machine can
independently recover about 1D6 /2 x 10% of the information. Total each
machines recovery percentage if they are connected again, that amount is the
total amount of data saved from corruption.
[This is totally bogus obviously, I think I was infected by reading too much
on NUMA machines. Hence this rule will probably be move to "NUMA-style"
architectures that WILL fail in such as case.]
CAPABILITIES
Once the clustered systems are properly connected then the systems combine
to form a "virtual" host on the highest rated system in the network. If
there are multiple machines with the same high rating then the player or
group setting up the cluster determines where the system forms. This
clustered host has the following capabilities.
[Well in all actuality I'd like to say that the single-system image is
really "nowhere" on the host but I don't want to feed the rampant bogosity
that systems can "gather" "free" clock cycles and memory to create
their own
virtual hosts (sorry about that UV host description natch)]
Core Rating
Total the Core Ratings of all the systems in the cluster and then subtract
the number of systems in the cluster squared. The total is the new machines
effective Core Rating.
For example, Asymmetric is clustering together three systems with Core
Ratings of 14, 10, and 20. Their new effective Core Rating will be (44-9=)
35. The new clustered host will form on the Core Rated 20 machine.
[This is an heavy-handed way of representing communications overhead and the
cluster services.]
System Load
The new effective Core Rating is used to calculate System Load for the
clustered host.
Active and Storage Memory
Total all of the clustered systems active and storage memories.
[See, sounds like NUMA more and more, what was I thinking!?]
Offline Storage
All of the machines offline storage devices are connected to the clustered
host, but are uniquely identified so users can choose a specific offline
storage device to connect to.
Environment and Persona Programs
Only one set of environment or persona programs can be ran on the machine.
If multiple copies are running on the machines when they are connected in
the cluster all but one must be unloaded.
[Environment program reads as "mainframe "persona" program.]
Location
When the clustered host is connected to the Matrix it's home LTG is the one
that contains the system it is running on, even if the other parts of the
cluster are in different LTGs.
[This part is going to be a bit tricky given the..odd...communications
structure of Shadowrun but confining them to an LTG should help.]
Kenneth
"On two occasions I have been asked, 'Pray, Mr. Babbage, if you put into the
machine wrong figures, will the right answers come out?' I am not able to
apprehend the kind of confusion of ideas that could provoke such a
question."
-- Charles Babbage
