The study to improve the probability of timely and errorfree redundant transmission of packet copies through aggregated channels, taking into account the in uence of the distributed queues organization and their dispatching based on the multiple access procedure is conducted. The simulation model of the data transmission system with aggregated channels and distributed queue with a marker access method is built. The boundaries of the e ectiveness of redundant services are determined taking into account delays in the implementation of multiple access to redundant channels.
High structural reliability [1{3], performance [
The study of the possibility of the average waiting time reducing in computer
implementations represented by multichannel queuing systems (QMS) as a
result of the requests replication with their independent execution by all or part
of unoccupied service channels was carried out in [
The study of the interaction of computer nodes organization through
redundant channels was considered in [
The object of the study is the system in which interaction between m computer nodes is performed through n aggregated channels (Figure 1).
Client 1 Client 2 Client 3 Client 4...Client m The feature of the systems under investigation organization is the presence of the queue, which is distributed across system nodes. The organization of this queue depends on the implementation of the interaction of nodes through redundant channels, including their redundant interaction.
We distinguish two limiting options for organizing distributed over nodes queues.
In option A, separate queues are organized at each node to access each of the m channels. Each request is placed in one of the queues with non-redundant service (for example, cyclically or in the least loaded queue). In case of redundant service with multiplicity k, k copies are created, each of which is entered into one of the n queues. When connecting computer nodes to channels (backbones) via network adapters, queues can be organized both in the computer node and in network adapters.
In the variant of queuing A, the n-channel data transmission system can
be represented as a set of n single-channel QMS [
With option B, a common queue at each computer node is organized to access all n channels. In case of non-redundant service after release (obtaining access privileges) of one of the n channels, a request at the beginning (output) of the queue is transmitted via it. In case of redundant service, a request that is at the head of the queue creates k copies, each of which is sent as it is released (granted access) to the k channels. When a computer node is connected (in which a queue common for all channels is organized) to n channels via n network adapters, k copies of the request from the top of the queue are put into k from the n adapter. With this queuing system, the data transmission system through the reserved channel is represented by multi-channel QMS with a common distributed queue.
The estimates of the average latency and timeliness of redundant servicing of the requests for packet transmission through redundant channels when presented by a combination of single-channel QMS can be based on the analytical models proposed in [16{18].
The e ectiveness of the backup service requests for the transfer of
packages through aggregated channels, when presenting the analyzed system as a
multi-channel QMS with a common queue, was studied in [
The most common is the random method of multiple access, but it does not
provide guaranteed access time (calculated for the worst case), which makes it
di cult to use it in real-time computer systems (including managing systems).
For real-time systems, it is advisable to use deterministic multiple access
methods, including the token [
Figure 2 shows the chart of the organization of a distributed queue using the token access method. Each channel has the token transmitted in accordance with a given logical ring of its transmission. The logical ring of the token transfer is organized for each of the n channels. If there is the request (packet) in the node's queue stored in the network adapter, then the adapter starts its transmission after token receiving. After the node completes (via the network adapter) the required transfers, the token is transmitted to the network adapter of the next node in the logical ring. If there are no packets to be sent in the adapter, then it is transmitted to the next node in the logical ring after token receiving. In the case of redundant transmissions when the common queue is organized in a computer node, from its beginning, k copies of the request are recorded in k of n network adapters, each packet is transmitted independently when the adapter receives a token included in the logical ring of the corresponding channel token. Due to the non-synchronous transmission of di erent channels tokens, the time di erence between the delivery of packets from one node through di erent channels is possible.
The aim of the work is to study the possibilities of increasing the probability of timely and error-free redundant transmissions of the packet copies through aggregated channels, taking into account the in uence of the distributed queues
Q u e u e 1
Q u e u e 2
Q u e u e 3 organization and their dispatching based on the multiple access procedure. The e ectiveness of the packet copies redundant transfers through aggregated channels is analyzed when presenting the system based on multi-channel QMS with a common queue distributed between the nodes.
The aim of the study is to resolve technical contradictions associated with the fact that redundant service: increases the likelihood of error-free packet delivery, but on the other hand leads to the load increase, which leads to the increase in the transmission time of packets through separate channels and, accordingly, reduce the probability of timely delivery.
As an indicator of e ciency, we use the multiplicative criterion of the form
M = P (t0 T ) [
The results of the study were obtained on the basis of the simulation model implemented in the AnyLogic 7 environment. 2
Figure 3 shows the simulation model of the data transmission system with redundant channels and the distributed queue organization based on the multiple access token method.
In this model, the sourcei blocks are the sources of i -th subscriber packets. Blocks spliti - create the required number of copies. The queuei blocks are endless queues of i -th subscriber packets. The holdi and delayWi blocks are responsible for the transfer of the token. The selectOutputIn and selectOutoutOuti blocks are the technical blocks responsible for the distribution of packets over communication channels. Blocks delayi - 8 identical communication channels. CheckErrori blocks - checks the package error-free delivery. Sinki blocks are responsible for keeping track of correctly delivered packages. The sinkFi blocks are responsible for keeping track of packets delivered with an error. SinkTi blocks are responsible for the packets accounting lost due to exceeding the maximum allowable time. 3
The simulation experiments were carried out with varying the total intensity of input streams with values = 1 ... 10000 1/s; m = 8 is the number of channels in the service system; L = 1 Mbit/s - the capacity of each of the communication channels; the probability of bit error in the channel B = 10 5; N = 1024 bits the average length of packets arriving in the system; maximum delivery time is t0 = 0,0002 s for the cases of redundant and non-redundant services.
Figures 4 and 5 show the dependencies of the service e ciency (timeliness and accuracy) estimated by the complex indicator M, without taking into account the costs of the multiple access procedure and taking them into account when using the marker access method, respectively. Transmission without reservation corresponds to curve 1 and with redundancy of multiplicity 2 - curve 2.
From the presented charts one can see the existence of the expediency of applying redundant services region at low values of the input request ow intensity. In this case, the multiple access procedure to communication channels loss leads to the reduction in the area of redundant services use expediency.
It is interesting to compare the simulation results without taking into account access to communication channels and with the token access method at di erent M = P(t0 -T )´10-5, s 7,9 7,85 7,8 7,75 7,7 7,65 7,6 7,85 7,8 7,75 7,7 7,65 7,6 0 1 2 3 4 5 6 7 8
9 10 l ´103, s-1 redundancy rates. Figures 6 and 7 show the charts of the e ciency in terms of M without redundancy and with redundancy ratio 2 respectively. Curves 1 and 2 correspond to the redundant transmissions e ectiveness in the case of taking into account and not taking into account losses of the token access method implementation. The di erence of e ciency with and without accounting the token access method implementation losses is represented by curve 3. 2 4 6 8
10-2 l ´103, s-1 The presented charts show the decrease in the e ciency of service when accounting the costs of providing access to communication channels.
The simulation model of the data transmission system through aggregated channels with the token method of multiple access is built. The boundaries of the expediency of the use of redundant service, taking into account multiple access to redundant channels are identi ed. The impact on the e ectiveness of redundant dispatching transmissions of queues distributed between computer nodes based on multiple access is evaluated.