Device and method for deterministic packet scheduling
Patent Information
- Application Number
- EP2022708051
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2022-02-08
- Publication Date
- 2024-11-20
AI Technical Summary
Existing deterministic packet scheduling algorithms face challenges in implementing latency bounds beyond 100μs to tens of ms under limited chip resources, with EDF being optimal but difficult to implement, and other algorithms either failing to provide guaranteed quality of service or requiring complex state variables and priority queues.
A deterministic packet scheduler using a rotating priority queue (RPQ) scheme with a plurality of FIFO queues, each associated with a common depth and unique delay bounds, where the processor serves the queue with the lowest delay bound and rotates the queues to maintain compliance with delay requirements, simplifying the scheduling process and accommodating various delay bounds.
This approach provides less complex, yet deterministic packet scheduling that effectively buffers delay-sensitive traffic up to 100μs and delay-tolerant traffic up to tens of ms, converging to optimal schedules while reducing implementation complexity and memory requirements.
Smart Images

Figure 1.1
Abstract
Description
[0001]DEVICE AND METHOD FOR DETERMINISTIC PACKET SCHEDULING Technical Field The present disclosure relates generally to the field of network / packet scheduling, and more particularly to a deterministic packet scheduler, a method of operating a deterministic packet scheduler, and a corresponding computer program. Background Art Deterministic packet scheduling of traffic requiring latency bounds beyond 100μs and up to tens of ms can be extremely challenging under limited chip resources such as total memory and / or total number of queues. Earliest deadline first (EDF) is an optimal scheduling algorithm in that a collection of independent jobs such as forwarding packets, each characterized by a deadline such as a latency bound, will be scheduled in a way that ensures all the jobs complete by their deadline. As such, EDF is able to perform deterministic packet scheduling in view of the aforementioned latency bounds, but is difficult to implement, too. Alternative scheduling algorithms such as RSCP, Jitter-VC, EDD, WFQ, etc. attempt to emulate EDF while reducing implementation complexity. However, they require to include a state variable (i.e., modification of packet headers), and / or a priority queue (or per-flow queuing). Other scheduling algorithms such as CBS, SP, (D)RR or PP are relatively easy to implement but fail to provide guaranteed quality of service (QoS). The rotating priority queue (RPQ) scheduling algorithm proposed by Liebeherr converges to EDF for small queue depths, but requires a high quantity of queues. Summary This disclosure and its solutions are based on the following further considerations. FIG.1 illustrates an example of a RPQ scheduler comprising a plurality of N first-in first-out (FIFO) queues (N=4 in the example of FIG.1). Each queue is associated with a common depth Δ (in a range of μs, for example) and a unique time- to-service TTSn (see right side of FIG.1). For descriptive reasons, the plurality of queues 111 is lined up along the vertical timeline from lowest to highest time-to-service TTSn. Each queue is further associated with a unique delay bound δn(see left side of FIG.1), so that arriving packets having respective packet delay requirements may be classified to those of the plurality of queues being associated with respective next larger delay bounds δn. For example, an arriving packet having a packet delay requirement between 0 and δn=1will be classified to the queue associated with the delay bound δn=1, an arriving packet having a packet delay requirement between δn=1and δn=2will be classified to the queue associated with the delay bound δn=2, and so forth. FIG.2 illustrates an example of a step of operation of the RPQ scheduler of FIG.1. First, a queue is served. Only one queue is served at a time, and for a service time period corresponding to the depth Δ. The served queue (indicated in FIG.2 as being emptied) is associated with a lowest delay bound δn=1= Δ (or equivalently, lowest time-to-service TTSn=1= 0). Subsequently, a rotation of queues is performed. For each of the plurality of queues but the served queue, the rotation involves being advanced (see arrows) in the line of queues in terms of a priority of being served. More specifically, the time-to- service TTSnof the respective queue is decreased by the service time period Δ, and the delay bound δnof the respective queue is set to correspond to the next smaller delay bound δn-1. For the empty served queue, the rotation involves being returned to the end of the line of queues. In other words, its delay bound δnis set to correspond to a maximum delay bound δn=4=Nof the plurality of queues, and its time-to-service TTSnto correspond to the set delay bound δnless the service time period Δ. (i.e., TTSn=4=N= δn=4=N– Δ). The sequence of serving and rotation according to FIG.2 is repeated over and over again. For example, deterministic packet scheduling in view of a 10ms latency bound requires a plurality of N = 10ms / Δ = 1000 (or 500) queues for queue depths of Δ = 10μs (or 20μs). It is an object to overcome these and other drawbacks by the features of the independent claims. In particular, it is an object to provide less complex, yet deterministic packet scheduling under hardware / chip limitations, and preferably converging to the optimal schedule. The foregoing and other objects are achieved by the features of the independent claims. Further implementation forms are apparent from the dependent claims, the description and the figures. According to a first aspect, a deterministic packet scheduler is provided. The scheduler comprises a plurality of first-in first-out, FIFO, queues, and a processor. The plurality of first-in first-out, FIFO, queues is respectively associated with a common depth, a unique delay bound, and a unique time-to- service. The processor is configured to serve a queue of the plurality of queues being associated with a lowest delay bound for a service time period corresponding to the depth. The processor is further configured, for each of the plurality of queues but the served queue, to decrease the time-to-service of the respective queue by the service time period; and if the decreased time-to-service corresponds to a next smaller delay bound of the plurality of queues, to set the delay bound of the respective queue to correspond to the next smaller delay bound. The processor is further configured, for the served queue, to set the delay bound to correspond to a predetermined delay bound being smaller than a maximum delay bound of the plurality of queues or to an alternative delay bound if the delay bound of any one of the plurality of queues corresponds to the predetermined delay bound, and to set the time-to-service to correspond to the set delay bound less the service time period. The deterministic packet scheduler buffers delay-sensitive traffic (up to a delay bound of 100μs, for example) in a set of FIFO queues rotating in accordance with RPQ scheduling, and delay-tolerant traffic (up to a significantly larger delay bound of tens of ms, for example) in an additional set of FIFO queues. The queues corresponding to the RPQ scheduler are generally rotated in accordance with an RPQ scheme, and the additional queues exceptionally interleave the rotating queues of the first set in accordance with a respective time-to-service TTSm / n, such that the buffered traffic is served in conformity with the applicable delay bounds. This provides less complex, yet deterministic packet scheduling. In a possible implementation form, the respective delay bounds may comprise respective integer multiples of the depth. Thereby, a rotation of the queues may be simplified., for example by integer calculation (i.e., decrementing counters instead of subtracting floating point values). In a possible implementation form, the respective delay bounds not exceeding the predetermined delay bound may differ pairwise by the depth. Thereby, the queues corresponding to the RPQ scheduler of FIG. 1 may still accommodate delay bound of, say, 100μs, for example. In a possible implementation form, the respective delay bounds exceeding the predetermined delay bound may differ pairwise by the respective integer multiples of the depth. Thereby, the additional queues with respect to the RPQ scheduler of FIG. 1 may accommodate significantly larger delay bounds of tens of ms, for example. In a possible implementation form, the alternative delay bound may comprise the maximum delay bound of the plurality of queues. Thereby, a maximum delay bound is covered regularly. In a possible implementation form, the alternative delay bound may comprise a packet delay requirement that is associated with a maximum packet arrival rate and a maximum packet arrival burst. Thereby, a most frequent delay bound is covered on-the-fly. In a possible implementation form, the processor may further be configured to initialize the plurality of queues. In a possible implementation form, the processor may further be configured to initialize the plurality of queues by, for each of the plurality of queues, setting the common depth, setting the unique delay bound; and setting the time-to-service to correspond to the set delay bound less the service time period. Thereby, a proper scheduling operation from the start is secured. In a possible implementation form, the processor may further be configured to classify packets having respective packet delay requirements to those of the plurality of queues being associated with respective next larger delay bounds. Thereby, packet classification in accordance with the extended delay bounds is achieved. In a possible implementation form, the processor may further be configured to serve the queue of the plurality of queues being associated with the lowest delay bound for the time period corresponding to the common depth by emptying the served queue completely given a predefined service rate. Thereby, the buffered traffic is served in conformity with the applicable delay bounds. In a possible implementation form, the plurality of queues may respectively be gate-controlled. In a possible implementation form, the plurality of queues may share a predefined maximum amount of on-chip memory. In a possible implementation form, the plurality of queues may share a predefined maximum count of on-chip queues. Thereby, scheduling operation takes into account realistic hardware / chip limitations. According to a second aspect, a method of operating a deterministic packet scheduler is provided. The scheduler comprises a plurality of first-in first-out, FIFO, queues. The plurality of queues is respectively associated with a common depth, a unique delay bound, and a unique time-to-service. The method comprises serving a queue of the plurality of queues being associated with a lowest delay bound for a service time period corresponding to the depth. The method comprises, for each of the plurality of queues but the served queue, decreasing the time-to-service of the respective queue by the service time period, and if the decreased time-to-service corresponds to a next smaller delay bound of the plurality of queues, setting the delay bound of the respective queue to correspond to the next smaller delay bound. The method further comprises, for the served queue, setting the delay bound to correspond to a predetermined delay bound being smaller than a maximum delay bound of the plurality of queues, or to an alternative delay bound if the delay bound of any one of the plurality of queues corresponds to the predetermined delay bound, and setting the time-to-service to correspond to the set delay bound less the service time period. In a possible implementation form, the method may be performed by the scheduler of the first aspect or any of its implementations. The technical effects and advantages described above in relation with the deterministic packet scheduler equally apply to the method of operating the same having corresponding features. According to a third aspect, a computer program is provided. The computer program comprises executable instructions, which when executed by a processor, cause the processor to perform the method of the second aspect or any of its implementations. Brief Description of Drawings The above-described aspects and implementations will now be explained with reference to the accompanying drawings, in which the same or similar reference numerals designate the same or similar elements. The drawings are to be regarded as being schematic representations, and elements illustrated in the drawings are not necessarily shown to scale. Rather, the various elements are represented such that their function and general purpose become apparent to those skilled in the art. FIG.1 illustrates an example of a rotating priority queue (RPQ) scheduler; FIG.2 illustrates a step of operation of the RPQ scheduler of FIG.1; FIG.3 illustrates a deterministic packet scheduler in accordance with the present disclosure; FIGs.4 – 9 illustrate steps of operation of the deterministic packet scheduler of FIG.3; and FIG.10 illustrates a method of operating the deterministic packet scheduler of FIG. 3 in accordance with the present disclosure. Detailed Descriptions of Drawings In the following description, reference is made to the accompanying drawings, which form part of the disclosure, and which show, by way of illustration, specific aspects of implementations or specific aspects in which implementations may be used. It is understood that implementations may be used in other aspects and comprise structural or logical changes not depicted in the figures. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present disclosure is defined by the appended claims. For instance, it is understood that a disclosure in connection with a described method may also hold true for a corresponding apparatus or system configured to perform the method and vice versa. For example, if one or a plurality of specific method steps are described, a corresponding device may include one or a plurality of units, e.g. functional units, to perform the described one or plurality of method steps (e.g. one unit performing the one or plurality of steps, or a plurality of units each performing one or more of the plurality of steps), even if such one or more units are not explicitly described or illustrated in the figures. On the other hand, for example, if a specific apparatus is described based on one or a plurality of units, e.g. functional units, a corresponding method may include one step to perform the functionality of the one or plurality of units (e.g. one step performing the functionality of the one or plurality of units, or a plurality of steps each performing the functionality of one or more of the plurality of units), even if such one or plurality of steps are not explicitly described or illustrated in the figures. Further, it is understood that the features of the various exemplary implementations and / or aspects described herein may be combined with each other, unless specifically noted otherwise. FIG.3 illustrates a deterministic packet scheduler 1 in accordance with the present disclosure. The scheduler 1 comprises a plurality of FIFO queues 111 and a processor 12. As used herein, a scheduler, network scheduler or packet scheduler may refer to an arbiter on a node in a packet-switched communication network managing the sequence of network packets in the queues of the protocol stack and network interface controller. As used herein, a queue may relate to a collection of entities that are maintained in a sequence and may be modified by adding or removing entities at either ends of the sequence. For example, the entities may relate to data packets (i.e., data traffic) buffered in an electronic memory. As used herein, FIFO or first-in first-out may relate to serving jobs, such as forwarding packets, buffered in a queue on a first-come, first-served basis. In the previous example, a FIFO queue may comprise a collection of data packets, which collection may be modified by adding arriving data packets at one end of the sequence or removing served data packets at the other end of the sequence. As shown in FIG. 3, the plurality of queues 111 may share a predefined maximum amount of on- chip memory 11, and may further share a predefined maximum count (i.e., quantity) of on-chip queues 111. The plurality of queues 111 may respectively be gate-controlled. As used herein, gate control may relate to controlling which queue is permitted to transmit at a given point in time. For example, only one queue 111 with a highest priority of being served (i.e., lowest time-to-service, see below) is allowed to transmit packets. The other queues 111 are closed to solely receive / buffer packets. The processor 12 may comprise a general-purpose microprocessor, a network processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), etc., for example. The scheduler 1 may be implemented as a system-on-a-chip (SoC) integrating circuit components such as the memory 11 and the processor 12. When executed by the processor 12, a computer program comprising executable instructions, causes the processor 12 to perform the method 2 of the second aspect or any of its implementations described in connection with FIG.10, and further specified in terms of device features in connection with FIGs. 4 – 9. FIGs.4 – 9 illustrate steps of operation of the deterministic packet scheduler 1 of FIG.3. The scheduler 1 comprises a plurality of N+M FIFO queues (N=4 and M=2 in the example of FIGs. 4 – 9) representing the plurality of N queues of the RPQ scheduler of FIG. 1 for applications with small latency and jitter (e.g., ~100μs) and a plurality of M additional queues for applications with larger latency demands (e.g., tens of ms), which approximate a number of α queues of the normal RPQ scheme. As used herein, latency may refer to delay. As used herein, jitter may refer to delay variance. A predetermined delay bound δn=4=Nrelates to the largest delay bound δn=4=Nof the plurality of N queues for applications with small latency and jitter (i.e., corresponding to the plurality of N queues of the RPQ scheduler of FIG.1). Each queue 111 is associated with a common (i.e., same) depth Δ (in a range of μs, such as 10μs or 20μs, for example) and a unique time-to-service TTSm / n(see right side of FIG.4). Again, the plurality of queues 111 is lined up along the vertically arranged timeline from lowest to highest time-to-service TTSm / nfor descriptive reasons. However, there is no need to physically move / rotate the queues 111. Those skilled in the art will appreciate that the plurality of queues 111 may be represented by respective data structures stored in the memory 11, between which data may be copied / moved selectively, or if the respective data structures are identified by unique indices or address pointers, then those “handles” may be used to logically move / rotate the queues 111. Each queue 111 is further associated with a unique delay bound δm / n(see left side of FIG.4), so that arriving packets having respective packet delay requirements may be classified to those of the plurality of queues 111 being associated with respective next larger delay bounds δm / n. For example, an arriving packet having a packet delay requirement between 0 and δn=1will be classified to the queue 111 associated with the delay bound δn=1, an arriving packet having a packet delay requirement between δn=1and δn=2will be classified to the queue 111 associated with the delay bound δn=2, an arriving packet having a packet delay requirement between δn=4=Nand δm=1will be classified to the queue associated with the delay bound δm=1, and so forth. In other words, the scheduler 1 of FIGs.4 – 9 is able to deal with packet delay requirements between 0 and δm=2=M. As can be seen, the plurality of M additional queues 111 is associated with delay bounds δm being significantly higher than the delay bounds δnof the plurality of N queues of the RPQ scheduler for tolerating more relaxed packet latency bounds, despite having the same queue depth Δ. That is to say, the respective delay bounds δm= δm-1+ αm·Δ = δm=1+ ∑2..m( αm·Δ) or δn= δn-1+ Δ = ∑1..n(Δ) may comprise respective integer multiples αm>1 or αn=1) of the depth Δ. More specifically, the respective delay bounds δn= δn-1+ Δ not exceeding the predetermined delay bound δn=N may differ pairwise by the depth Δ, and the respective delay bounds δm= δm-1+ αm·Δ exceeding the predetermined delay bound δn=Nmay differ pairwise by the respective integer multiples αmof the depth Δ. In the example of FIGs.4 – 9, each of the plurality of N=4 queues corresponding to RPQ scheduling merely deals with packet delay requirements extending over a range of Δ, whereas each of the plurality of M=2 additional queues covers packet delay requirements extending over a range of αΔ (see left side of FIG.4). Of note, α may generally differ for different indices m. FIGs.5 – 7 illustratessimilar steps of operation of the deterministic packet scheduler 1 of FIG.3. In this connection, reference will be made to device features which relate to corresponding method steps illustrated in FIG.10. The processor 12 is configured to serve 23 (see FIG.10) a queue 111’ of the plurality of queues 111. Only that queue 111’ is served at a time, and for a service time period corresponding to the depth Δ. The served queue 111’ (indicated in FIGs.5 – 7 as being emptied) is associated with a lowest delay bound δn=1= Δ (or equivalently, lowest time-to-service TTSn=1= 0). Subsequently, a rotation of the queues 111 is performed. For each of the plurality of queues 111 but the served queue 111’, the processor 12 is configured to decrease 24A (see FIG.10) the time-to-service TTSm / nof the respective queue 111 by the service time period Δ, and is further configured to set 25A (see FIG. 10) the delay bound δm / nof the respective queue 111 to correspond to the next smaller delay bound δm-1 / n-1if the decreased time-to-service TTSm / ncorresponds to that next smaller delay bound δm-1 / n-1. In FIGs.5 – 7, the time-to-service TTSm / nof each of the plurality of queues 111 is decreased 24A (see FIG.10) by the service time period Δ, and the delay bound δnof three of the plurality of N=4 queues 111 is set 25A (see FIG. 10) to correspond to the next smaller delay bound δn-1, as their decreased time-to-service TTSncorresponds to that next smaller delay bound δn-1. In other words, each of the plurality of queues 111 but the served queue 111’ advances in the line of queues 111 in terms of a priority of being served. For the served queue 111’, the processor 12 is configured to set 25B (see FIG.10) the delay bound δm / nto correspond to a predetermined delay bound δn=N being smaller than a maximum delay bound δm=Mof the plurality of queues 111, or to an alternative delay bound δm ≤Mif the delay bound δm / nof any one of the plurality of queues 111 corresponds to the predetermined delay bound δn=N, and to set 24B (see FIG.10) the time-to-service TTSm / nto correspond to the set delay bound δm / nless the service time period Δ. On the one hand, the alternative delay bound δm ≤Mmay comprise the maximum delay bound δm=Mof the plurality of queues 111. On the other hand, the alternative delay bound δm ≤Mmay comprise a packet delay requirement that is associated with a maximum packet arrival rate and a maximum packet arrival burst. In other words, a served queue 111’ may be returned at the end of the line of queues 111, as well as into the line of queues 111 in accordance with the most frequently observed packet delay requirements. According to exemplary simulation scenarios, both options perform comparably well. In FIGs. 5 – 7, none of the delay bounds δm / nof the plurality of queues 111 corresponds to the predetermined delay bound δn=4=N. As such, the served queue 111’ is returned to the end of the line of the queues 111 of RPQ scheduling by setting 25A (see FIG.10) the delay bound δm / nto correspond to the predetermined delay bound δn=4=Nand setting 24B (see FIG.10) the time-to-service TTSm / nto correspond to the set delay bound δm / nless the service time period Δ (i.e., TTSn=4=N= δn=4=N– Δ). In the example of FIGs. 5 – 7, the steps of operation result in five queues 111 of the plurality of queues 111 being lined up back-to-back in terms of their respective times-to-service TTSm / n. FIG.8 illustrates a further step of operation of the deterministic packet scheduler 1 of FIG.3. Again, the processor 12 is configured to serve 23 (see FIG.10) the queue 111’. Only that queue 111’ is served at a time, and for a service time period corresponding to the depth Δ. The served queue 111’ (indicated in FIG. 5 as being emptied) is associated with a lowest delay bound δn=1= Δ (or equivalently, lowest time-to-service TTSn=1= 0). Subsequently, a rotation of the queues 111 is performed. For each of the plurality of queues 111 but the served queue 111’, the processor 12 is configured to decrease 24A (see FIG.10) the time-to-service TTSm / nof the respective queue 111 by the service time period Δ, and is further configured to set 25A (see FIG. 10) the delay bound δm / nof the respective queue 111 to correspond to the next smaller delay bound δm-1 / n-1if the decreased time-to-service TTSm / ncorresponds to that next smaller delay bound δm-1 / n-1. In FIG. 8, the time-to-service TTSm / nof each of the plurality of queues 111 is decreased 24A (see FIG. 10) by the service time period Δ, and the delay bound δnof four of the five queues 111 being lined up back-to-back is set 25A (see FIG.10) to correspond to the next smaller delay bound δn-1, as their decreased time-to-service TTSncorresponds to that next smaller delay bound δn-1. In other words, each of the plurality of queues 111 but the served queue 111’ advances in the line of queues 111 in terms of a priority of being served. For the served queue 111’, the processor 12 is configured to set 25B (see FIG.10) the delay bound δm / nto correspond to a predetermined delay bound δn=Nbeing smaller than a maximum delay bound δm=Mof the plurality of queues 111, or to an alternative delay bound δm ≤Mif the delay bound δm / nof any one of the plurality of queues 111 corresponds to the predetermined delay bound δn=N, and to set 24B (see FIG.10) the time-to-service TTSm / nto correspond to the set delay bound δm / nless the service time period Δ. In FIG. 8, the delay bound δn=4=Nof one (shaded) of the plurality of queues 111 does in fact correspond to the predetermined delay bound δn=4=N. In other words, the shaded queue formerly indexed by m=1 (see FIG.7, and now indexed by n=4=N, see FIG.8) has an updated delay bound δn=4=N that corresponds to the predetermined delay bound δn=4=N. As such, the served queue 111’ is returned into the line of the queues 111 by setting 25A (see FIG.10) the delay bound δnto correspond to the alternative delay bound δm ≤M(in the example of FIG.8, to the maximum delay bound δm=M, so that the served queue 111’ is returned to the end of the line of the queues 111) and setting 24B (see FIG. 10) the time-to-service TTSm / nto correspond to the set delay bound δm / nless the service time period Δ (i.e., TTSm=2=M= δm=2=M– Δ). FIG.9 illustrates a further step of operation of the deterministic packet scheduler 1 of FIG.3 similar to FIG.5. The queues 111 corresponding to the RPQ scheduler are continually rotated in accordance with an RPQ scheme, and the additional queues (shaded in FIGs.4 – 9) exceptionally interleave the rotating queues of the RPQ set in accordance with a respective time-to-service TTSm / n, such that the buffered traffic is served in conformity with the applicable delay bounds. This provides less complex, yet deterministic packet scheduling of traffic requiring latency bounds beyond 100μs and up to tens of ms under limited chip resources such as on-chip memory and / or on- chip queues. Moreover, a schedulable region converges to EDF with decreased queue depth Δ. According to exemplary simulation scenarios, latency bounds may be increased into the millisecond regime. For example, accommodating 2ms or 5ms latency bounds with 25 queues requires ~62.5% or ~25% of the RPQ memory requirement. According to similar simulation scenarios, a queueing performance may be improved by increasing the number M of additional queues. For example, when increasing the total number of queues from 25 to 55, performance degradation is reduced from 1.89x of RPQ (95 percentile) to 1.38x of RPQ with respect to the average value, from 2.21 to no degradation with respect to the 95 percentile, and from 6.46x to 4.15x regarding a maximum service time deviation compared to EDF. FIG. 10 illustrates a method 2 of operating the deterministic packet scheduler 1 of FIG. 3 in accordance with the present disclosure. The scheduler 1 comprises a processor 12 as already mentioned above. The processor 12 may be configured to initialize 21 the plurality of queues 111, in particular by setting 211 the common depth Δ, setting 212 the unique delay bound δm / n; and setting 213 the time- to-service TTSm / nto correspond to the set delay bound δm / nless the service time period Δ, for each of the plurality of queues 111. When setting 212 the unique delay bound δm / nfor each of the plurality of queues 111, the respective delay bounds δm= δm-1+ ^m·Δ = δm=1+ ∑2..m( αm·Δ) or δn= δn-1+ Δ = ∑1..n(Δ) may comprise respective integer multiples ^m>1 or ^n=1) of the depth Δ. More specifically, the respective delay bounds δn= δn-1+ Δ not exceeding the predetermined delay bound δn=N may differ pairwise by the depth Δ, and the respective delay bounds δm= δm-1+ αm·Δ exceeding the predetermined delay bound δn=Nmay differ pairwise by the respective integer multiples αmof the depth Δ. The processor 12 may further be configured to classify 22 packets having respective packet delay requirements to those of the plurality of queues 111 being associated with respective next larger delay bounds δm / n. The method 2 comprises serving 23 a queue 111’ of the plurality of queues 111 being associated with a lowest delay bound δn=1for a service time period corresponding to the depth Δ (likewise, the processor 12 may be configured to serve 23 the queue 111’), preferably by emptying 231 the served queue 111’ completely given a predefined service rate C. As an operational assumption, the served queue 111’ is always emptied, and the queues 111 do not overflow (i.e., arriving traffic volume per queue 111 does not exceed the product of the queue depth Δ and the predefined service rate C (outgoing bandwidth). As shown on the left side of FIG. 10, for each of the plurality of queues 111 but the served queue 111’ the method 2 comprises decreasing 24A the time-to-service TTSm / nof the respective queue by the service time period Δ, and if the decreased time-to-service TTSm / ncorresponds to a next smaller delay bound δm-1 / n-1of the plurality of queues 111, setting 25A the delay bound δm / nof the respective queue 111 to correspond to the next smaller delay bound δm-1 / n-1. As can be seen on the right side of FIG.10, for the served queue 111’ the method 2 further comprises setting 25B the delay bound δnto correspond to a predetermined delay bound δn=Nbeing smaller than a maximum delay bound δm=Mof the plurality of queues 111, or to an alternative delay bound δm ≤Mif the delay bound δnof any one of the plurality of queues 111 corresponds to the predetermined delay bound δn=N, and setting 24B the time-to-service TTSm / nto correspond to the set delay bound δm / nless the service time period Δ. The method 2 may be performed by the scheduler 1 of the first aspect or any of its implementations, by a computer program comprising executable instructions which, when executed by the processor 12, causes the processor 12 to perform the method 2 of the second aspect or any of its implementations. The present disclosure has been described in conjunction with various examples as well as implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed matter, from the studies of the drawings, this disclosure and the independent claims. In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation. A computer program may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.
Claims
Claims 1. A deterministic packet scheduler (1), comprising a plurality (N+M) of first-in first-out, FIFO, queues (111), respectively being associated with a common depth (Δ), a unique delay bound (δm / n), and a unique time-to-service (TTSm / n); and a processor (12), configured to - serve (23) a queue (111’) of the plurality of queues (111) being associated with a lowest delay bound (δn=1) for a service time period corresponding to the depth (Δ); - for each of the plurality of queues (111) but the served queue (111’), o decrease (24A) the time-to-service (TTSm / n) of the respective queue (111) by the service time period (Δ); and o if the decreased time-to-service (TTSm / n) corresponds to a next smaller delay bound (δm-1 / n-1) of the plurality of queues (111), set (25A) the delay bound (δm / n) of the respective queue (111) to correspond to the next smaller delay bound (δm-1 / n-1); and - for the served queue (111’), o set (25B) the delay bound (δn) to correspond to a predetermined delay bound (δn=N) being smaller than a maximum delay bound (δm=M) of the plurality of queues (111), or to an alternative delay bound (δm ≤M) if the delay bound (δn) of any one of the plurality of queues (111) corresponds to the predetermined delay bound (δn=N); and o set (24B) the time-to-service (TTSm / n) to correspond to the set delay bound (δm / n) less the service time period (Δ).
2. The scheduler (1) of claim 1, the respective delay bounds (δm= δm-1+ αm·Δ = δm=1+ ∑2..m( αm·Δ); δn= δn-1+ Δ = ∑1..n(Δ)) comprising respective integer multiples ( αm; 1) of the depth (Δ).
3. The scheduler (1) of claim 2, the respective delay bounds (δn= δn-1+ Δ) not exceeding the predetermined delay bound (δn=N) differ pairwise by the depth (Δ).
4. The scheduler (1) of claim 2 or claim 3, the respective delay bounds (δm= δm-1+ αm·Δ) exceeding the predetermined delay bound (δn=N) differ pairwise by the respective integer multiples ( αm) of the depth (Δ).
5. The scheduler (1) of any one of the claims 1 – 4, the alternative delay bound (δm ≤M) comprising the maximum delay bound (δm=M) of the plurality of queues (111).
6. The scheduler (1) of any one of the claims 1 – 4, the alternative delay bound (δm ≤M) comprising a packet delay requirement that is associated with a maximum packet arrival rate and a maximum packet arrival burst.
7. The scheduler (1) of any one of the preceding claims, the processor (12) further being configured to - initialize (21) the plurality of queues (111).
8. The scheduler (1) of claim 7, the processor (12) further being configured to initialize (21) the plurality of queues (111) by, for each of the plurality of queues (111), - setting (2111) the common depth (Δ); - setting (212) the unique delay bound (δm / n); and - setting (213) the time-to-service (TTSm / n) to correspond to the set delay bound (δm / n) less the service time period (Δ).
9. The scheduler (1) of any one of the preceding claims, the processor (12) further being configured to - classify (22) packets having respective packet delay requirements to those of the plurality of queues (111) being associated with respective next larger delay bounds (δm / n).
10. The scheduler (1) of any one of the preceding claimsthe processor (12) further being configured to serve (23) the queue (111’) of the plurality of queues (111) being associated with the lowest delay bound (δn=1) for the time period corresponding to the common depth (Δ) by - emptying (231) the served queue (111’) completely given a predefined service rate (C).
11. The scheduler (1) of any one of the preceding claims, the plurality of queues (111) respectively being gate-controlled.
12. The scheduler (1) of any one of the preceding claims, the plurality of queues (111) sharing a predefined maximum amount of on-chip memory (11).
13. The scheduler (1) of any one of the preceding claims, the plurality of queues (111) sharing a predefined maximum count of on-chip queues (111).
14. A method (2) of operating a deterministic packet scheduler (1), the scheduler (1) comprising a plurality (N+M) of first-in first-out, FIFO, queues (111), respectively being associated with a common depth (Δ), a unique delay bound (δm / n), and a unique time-to-service (TTSm / n); the method (2) comprising - serving (23) a queue (111’) of the plurality of queues (111) being associated with a lowest delay bound (δn=1) for a service time period corresponding to the depth (Δ); - for each of the plurality of queues (111) but the served queue (111’), o decreasing (24A) the time-to-service (TTSm / n) of the respective queue by the service time period (Δ); and o if the decreased time-to-service (TTSm / n) corresponds to a next smaller delay bound (δm-1 / n-1) of the plurality of queues (111), setting (25A) the delay bound (δm / n) of the respective queue (111) to correspond to the next smaller delay bound (δm-1 / n-1); and - for the served queue (111’),o setting (25B) the delay bound (δn) to correspond to a predetermined delay bound (δn=N) being smaller than a maximum delay bound (δm=M) of the plurality of queues (111), or to an alternative delay bound (δm ≤M) if the delay bound (δn) of any one of the plurality of queues (111) corresponds to the predetermined delay bound (δn=N); and o setting (24B) the time-to-service (TTSm / n) to correspond to the set delay bound (δm / n) less the service time period (Δ).
15. The method (2) of claim 14, being performed by the scheduler (1) of any one of the claims 1 – 13.
16. A computer program comprising executable instructions, which when executed by a processor (12), cause the processor (12) to perform the method (2) of claim 14 or claim 15.
