processed even before real-time package if the priority and timing for this package
become critical.
Router queues are organized so far as Fig. 19.8a qualitatively illustrates. The
theory of this kind of queues is well developed by Kleinrock [11]. This theory
estimates the position in a queue for the process with different states.
What is not considered in [11] is that what is proposed here: queued element
(package in a queue) in real world should change its priority:—from very low to
highest possible. Our desperation principle is just about it. Priority for RT packages
is growing as long as its time left to destination is shrinking; Eq. 19.14 defined a
so-to-say “a probability” of loss is growing (in other words, desperation goes
beyond the chart).
In terms of implementation we might consider that packages should be described
by two priorities—a static as given initially and dynamic—that is changing along
the package journey in the network and a good router should handle and deal with
both of them. We can write it symbolically:
While P "; L ! 0
ð19:14Þ
In plain English: when desperation of the package grows a position in the queue
should be changed toward the output, leaving the queue (Fig. 19.8b).
As it was mentioned already, an “inside job” should be seriously revised to assist
implementation of a principle of desperation control at the router level. Let us check
what is possible here.
We need to elaborate this rule a bit further and see how to see it and implement
it: Fig. 19.8c.
DL stands for desperation level of a package for the whole journey. Inside an i-th
router delay for this package consists of input queueing delay I q , processing delay
(routing processing) RP i and Output queueing delay O q . All three might and should
be revised in structure and overheads accordingly desperation level of a package.
Note that routing delay RP i is squeezed as shown in previous sections of this
work. Further delays I q and O q should be managed as Fig. 19.8b shows—moving a
packages across the queuing accordingly desperation level.
Obviously that performance gain is scaled at order of magnitude, but for explicit
results the whole structure of desperation control scheme should be analyzed using
network simulation tool.
19.11 Aftermath—Instead of Conclusion
• We have shown how “level of desperation” works. We have shown also that
desperation control in combination with vector heuristic perform by far better
than known routing algorithms.
290
19 Distributed Systems: Resilience, Desperation
become critical.
Router queues are organized so far as Fig. 19.8a qualitatively illustrates. The
theory of this kind of queues is well developed by Kleinrock [11]. This theory
estimates the position in a queue for the process with different states.
What is not considered in [11] is that what is proposed here: queued element
(package in a queue) in real world should change its priority:—from very low to
highest possible. Our desperation principle is just about it. Priority for RT packages
is growing as long as its time left to destination is shrinking; Eq. 19.14 defined a
so-to-say “a probability” of loss is growing (in other words, desperation goes
beyond the chart).
In terms of implementation we might consider that packages should be described
by two priorities—a static as given initially and dynamic—that is changing along
the package journey in the network and a good router should handle and deal with
both of them. We can write it symbolically:
While P "; L ! 0
ð19:14Þ
In plain English: when desperation of the package grows a position in the queue
should be changed toward the output, leaving the queue (Fig. 19.8b).
As it was mentioned already, an “inside job” should be seriously revised to assist
implementation of a principle of desperation control at the router level. Let us check
what is possible here.
We need to elaborate this rule a bit further and see how to see it and implement
it: Fig. 19.8c.
DL stands for desperation level of a package for the whole journey. Inside an i-th
router delay for this package consists of input queueing delay I q , processing delay
(routing processing) RP i and Output queueing delay O q . All three might and should
be revised in structure and overheads accordingly desperation level of a package.
Note that routing delay RP i is squeezed as shown in previous sections of this
work. Further delays I q and O q should be managed as Fig. 19.8b shows—moving a
packages across the queuing accordingly desperation level.
Obviously that performance gain is scaled at order of magnitude, but for explicit
results the whole structure of desperation control scheme should be analyzed using
network simulation tool.
19.11 Aftermath—Instead of Conclusion
• We have shown how “level of desperation” works. We have shown also that
desperation control in combination with vector heuristic perform by far better
than known routing algorithms.
290
19 Distributed Systems: Resilience, Desperation