CN102394829A - Reliability demand-based arbitration method in network on chip - Google Patents

Reliability demand-based arbitration method in network on chip Download PDF

Info

Publication number
CN102394829A
CN102394829A CN 201110358841 CN201110358841A CN102394829A CN 102394829 A CN102394829 A CN 102394829A CN 201110358841 CN201110358841 CN 201110358841 CN 201110358841 A CN201110358841 A CN 201110358841A CN 102394829 A CN102394829 A CN 102394829A
Authority
CN
Grant status
Application
Patent type
Prior art keywords
arbitration
reliability
information
flit
network
Prior art date
Application number
CN 201110358841
Other languages
Chinese (zh)
Inventor
付宇卓
刘婷
焦佳佳
Original Assignee
上海交通大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date

Links

Abstract

The invention relates to a reliability demand-based arbitration method in a network on chip (NoC). The method is characterized in that the method comprises the following steps that: (1), reliability information is arranged for all flits of a data packet; and (2), responses are carried out on arbitration requests of the flits according to the reliability information. According to the invention, arbitration mechanisms of a virtual channel allocator (VCA) and a switch allocator (SA) in an NoC can be fully optimized; application priority transmission of high reliability demands can be ensured; communication delay in the network can be reduced and an occurrence probability of a soft error during the transmission process can also be reduced. Meanwhile, a starvation problem can be solved by employing arbitration historical information; and effectiveness and high efficiency of the arbitration method with whole reliable sensing can be ensured; therefore, the method has an important practical value.

Description

片上互连网络中基于可靠性需求的仲裁方法 Arbitration reliability requirements based on chip interconnect network

技术领域 FIELD

[0001] 本发明涉及片上互连网络(Network on Chip, NoC)路由器微结构中的仲裁方法,尤其涉及基于应用需求可靠性的仲裁方法。 [0001] The present invention relates to an on-chip interconnect network (Network on Chip, NoC) router microstructures arbitration method, in particular, relates to a method of arbitration based on the application requirements of reliability.

背景技术 Background technique

[0002] 在众核(many-core)主导的片上并行计算系统中,NoC作为基础的通信网络连接各计算和存储单元,I/O设备等,用以传输系统处理的数据和控制信号。 [0002] The parallel computing system on a many-core (many-core) leading chip, as a basis for the NoC communication network connecting the computing and storage unit, I / O devices, etc., for data and control signals processed by the system. 其中,仲裁策略是影响NoC的路由微结构的性能,可靠性和功耗的重要因素之一。 Among them, the arbitration strategy is an important factor affecting the performance of NoC routing microstructure, reliability and power consumption. 因此,有必要设计和实现高效的仲裁系统来保证NoC通信质量,进而提高整个众核系统的工作效率和可靠性。 Therefore, it is necessary to design and implement an efficient arbitration system to ensure the NoC communication quality, thereby improving the efficiency and reliability of the whole many-core system.

[0003] 一般来说,典型的五端口的NoC微结构如图1所示,包括东南西北四个端口和邻居节点路由器相连,还有一个端口和本地处理单元链接,且五个端口由一个切5的交叉开关提供相互通路。 [0003] In general, the typical microstructures NoC five-port 1, including the four corners of neighbor nodes connected to the router and a port, and a local processing unit and a link port, a port and a tangential five 5 crossbar switch provides each channel.

[0004] 图1所示的现有NoC微结构主要包括五个模块:输入,路由计算(RoutingComputation,简称RC),虚通道申请(Virtual Channel Allocator,简称VCA),交叉开关申请(Switch Allocator,简称SA)和输出,其中输入主要有多个虚通道(Virtual Channel,简称VC)构成。 NoC microstructure shown in the prior [0004] FIG. 1 mainly includes five modules: an input, route calculation (RoutingComputation, referred to as the RC), virtual channel applications (Virtual Channel Allocator, referred to as the VCA), application crossbar (Switch Allocator, Acronym SA) and an output, the input of which there are a plurality of virtual channels (virtual channel, referred to as VC) configuration. 一个数据包(packet)通常被化分为一个头微片(head flit),多个数据微片(data flit)和一个尾微片(tail flit)串行传输。 A data packet (Packet) is typically divided into a header flit of (head flit), a plurality of data flits (data flit) and a tail flit (tail flit) serial transmission.

[0005] 现有的片上互连网络的路由多采用虫洞流控制机制,因此只有头微片经过RC和VCA,所有微片进行SA操作。 [0005] The existing routes on-chip interconnect network using more wormhole flow control mechanism, so that only through the header flit and the VCA RC, all flit for SA operation. 由于虚通道和交叉开关作为有限的片上资源会被来自不同端口的微片竞争,因此NoC仲裁系统主要是面向VCA和SA的设计和实现。 Since the virtual channel and a crossbar flit chip competition limited resources are from a different port, thus NoC arbitration system is mainly for design and implementation of SA and VCA.

[0006] 传统的仲裁主要着眼于公平性和性能的优化,然而随着工艺尺寸的缩小,供电电压的降低,串扰等造成的信号翻转现象,即软错误(soft error),使得NoC面临可靠性挑战。 [0006] a conventional arbitration and fairness mainly focused on optimizing performance, but with flipped signal process geometries shrink, supply voltage is reduced, crosstalk caused, i.e., a soft error (soft error), so that the face reliability NoC challenge. 而仲裁策略直接影响了NoC通信质量,尤其是一个应用中存在不同等级的可靠性需求。 The NoC arbitration policy directly affects the communication quality, especially in the presence of different levels of reliability requirements of an application. 例如,H. 264包括A、B、C三种类型数据分割,其中A型分割是头信息划分,包括宏块类型、量化参数和运动矢量,这个信息是最重要的。 For example, H. 264 including A, B, C are three types of divided data, wherein A is a divided type divided header information, including a macroblock type, a quantization parameter and the motion vector, this information is the most important. B型分割是帧内信息划分,帧内信息可以阻止错误的传播,该型数据分割要求给定分片的A型分割有效,相对于帧间信息,帧内信息能更好地阻止漂移效应,因此它比帧间分割更为重要。 B is an intra type segmenting information is divided, the intra information can prevent propagation of errors, this type of data division type A request to a given slice division effective, information with respect to inter, intra information to better prevent drift effect, Therefore it is more important than inter-division. C型分割是帧间信息划分,包括帧间CBI^s帧间系数,一般情况下它是编码分片的最大分区。 C-segmentation is divided into information frames, including inter CBI ^ s inter coefficients, it is generally the largest partitions of the encoded slice. 帧间分割是最不重要的,它的使用要求A型分割有效,由此不同重要性决定了不同级别的可靠性需求。 Inter segmentation is the most important, and it requires the use of A-type split effective, which determines the importance of the different needs of different levels of reliability.

[0007] 所以,尝试采用一些可靠性感知信息指导仲裁系统的设计和实现是必要的。 [0007] Therefore, try to use some design and implementation is necessary reliability-aware information to guide arbitration system. 然而,业界尚未提出在片上互连网络中基于应用可靠性需求的仲裁方法。 However, the industry has not yet proposed arbitration application reliability requirements based on-chip interconnect network.

发明内容 SUMMARY

[0008] 本发明要解决的技术问题是考虑基于众核系统的应用中不同类型数据不同等级可靠性需求的问题,实现高效的可靠性感知仲裁设计,在尽可能减少性能损失和额外开销的条件下,满足应用在NoC中的通信可靠性需求。 [0008] The present invention to solve the technical problem is to consider the issue based on many-core systems applications in different levels of different types of data reliability requirements, reliability perception arbitration efficient design, performance loss and minimize overhead conditions lower, to meet the reliability requirements of the application in the communication of NoC. [0009] 因此,本发明的目的是提供一种片上互连网络的基于可靠性需求的仲裁方法,其特征在于,包括如下步骤:(1)为数据包的各微片设置可靠性信息;(¾根据所述可靠性信息对所述微片的仲裁请求进行响应。 [0009] Accordingly, an object of the present invention is to provide an on-chip interconnect network arbitration method based on reliability requirements, characterized by comprising the steps of: (1) the reliability of information provided to the respective flits of the packet; ( ¾ reliability information according to the arbitration of the response request flit.

[0010] 一些实施例中,所述可靠性信息包括历史信息,所述历史信息记录仲裁成功或失败的动态改变。 History information [0010] In some embodiments, the reliability information includes a history information recording arbitration dynamically changing success or failure.

[0011] 一些实施例中,所述历史信息由计数器表示,其中当仲裁成功,则所述计数器的值重置为零,否则进行加1取模操作。 [0011] In some embodiments, the history information is represented by a counter, wherein when the arbitration is successful, then the counter value is reset to zero, otherwise are incremented modulo operation.

[0012] 一些实施例中,所述可靠性信息还包括可靠性需求信息,并且所述可靠性需求信息包括高可靠性需求和一般可靠性需求。 [0012] In some embodiments, the reliability information further includes information reliability requirements, and the high reliability requirements of reliability information includes general requirements and reliability requirements.

[0013] 一些实施例中,所述可靠性需求信息用η比特的标志位来记录,并且所述标志位设在所述微片中。 [0013] In some embodiments, the reliability information η demand flag bit is recorded, and the flag bit is set in the flit.

[0014] 一些实施例中,若多个所述微片的所述历史信息相同,则根据所述可靠性需求对所述历史信息相同的多个微片进行仲裁。 [0014] In some embodiments, if a plurality of pieces of the same micro-history information, arbitration is performed in accordance with the same reliability requirements for the plurality of history information flits.

[0015] 一些实施例中,若多个所述微片的所述历史信息相同,则根据所述可靠性需求对所述历史信息相同的多个微片进行仲裁。 [0015] In some embodiments, if a plurality of pieces of the same micro-history information, arbitration is performed in accordance with the same reliability requirements for the plurality of history information flits.

[0016] 一些实施例中,若多个所述微片的所述可靠性需求相同,则使用轮询仲裁进行优先级设置和选择。 [0016] In some embodiments, a plurality of the same if the flit reliability requirements, the polling arbitration priority setting and selection.

[0017] 应用本发明可以充分优化片上互连网络中VCA和SA的仲裁机制,能够保证高可靠性需求的应用优先传输,减少在网络中的通信延迟,减少传输过程中软错误发生概率。 [0017] Application of the present invention can fully optimize the on-chip interconnect network VCA SA and arbitration mechanisms, it is possible to ensure the transmission of high priority application reliability requirements, to reduce communication delay in the network, reduce the probability of a soft error occurs during transmission. 同时还采用仲裁历史信息解决饿死问题,保证整个可靠感知的仲裁方法的有效和高效,具有重要实用价值。 It also uses historical information to solve starvation problem arbitration, the arbitration process to ensure the overall effective and efficient and reliable perception has important practical value.

[0018] 结合附图,根据下文的通过示例说明本发明主旨的描述可清楚本发明的其他方面和优点。 [0018] conjunction with the accompanying drawings, the gist of the present invention DESCRIPTION Other aspects and advantages may be apparent according to the invention by way of example below.

附图说明 BRIEF DESCRIPTION

[0019] 结合附图,通过下文的述详细说明,可更清楚地理解本发明的上述及其他特征和优点,其中: [0019] conjunction with the accompanying drawings, described by the following detailed description, may be more clearly understood and other features and advantages of the present invention, wherein:

[0020] 图1示出了现有NoC的微结构; [0020] FIG 1 illustrates a conventional microstructure of the NoC;

[0021] 图2为示出根据本发明实施例的仲裁方法的流程图。 [0021] FIG 2 is a flowchart illustrating an embodiment of the arbitration method of the present invention.

具体实施方式 detailed description

[0022] 参见示出本发明实施例的附图,下文将更详细地描述本发明。 [0022] Referring to the accompanying drawings showing an embodiment of the present invention, the present invention is described in more detail below. 然而,本发明可以以许多不同形式实现,并且不应解释为受在此提出之实施例的限制。 However, the present invention may be embodied in many different forms and should not be construed as limiting, proposed by this embodiment. 相反,提出这些实施例是为了达成充分及完整公开,并且使本技术领域的技术人员完全了解本发明的范围。 Instead, these embodiments are proposed in order to achieve full and complete disclosure, and the person skilled in the art fully understand the scope of the invention. 这些附图中,为清楚起见,可能放大了层及区域的尺寸及相对尺寸。 In these figures, for clarity, it may be exaggerated sizes and relative sizes of layers and regions.

[0023] 以通用的NoC路由器结构为例,描述根据本发明的仲裁方法。 [0023] In general NoC router structure as an example, the method described arbitration according to the invention.

[0024] 实施例中,采用了名词虫洞交换机制、路由计算RC、虚通道的申请仲裁器VCA及交叉开关的申请仲裁器SA。 [0024] In an embodiment, the term wormhole exchange mechanism employed, the route calculation application arbiter VCA RC, and virtual channel arbiter CROSS switch SA.

[0025] 在虫洞交换机制下,微片作为最小的通信单元,路由器收到报头微片后不等待包尾到达就向下级路由传递,在网络中以流水方式传送。 Without waiting to reach the end of the packet is transmitted to the rear lower routing [0025] In wormhole exchange mechanism, as a minimum flit communication unit, the router receives the header flit, in a pipeline manner in the transport network. 当任一微片发生阻塞时,整个传送路径都被占用。 When a microchip of any obstruction occurs, the entire transmission path is occupied.

[0026] 路由计算RC系指根据网络中微片中的源-目的信息用以计算路由路径,可以基于路由表配置,也可以采用定制的逻辑模块实现下一跳路由节点的计算。 [0026] The route computing means RC network source micro sheet - to calculate the routing path based on the routing table configuration information object may be achieved using a custom logic calculates the next hop routing node.

[0027] 虚通道的申请仲裁器VCA系指为了解决虫洞交换机制的阻塞问题,通常采用增加虚通道的方法。 [0027] Application of the virtual channel arbiter means to solve the blocking problem VCA wormhole exchange mechanism, usually a method to increase virtual channels. 基本思想是在每个物理通道上设置多个缓冲区,即虚通道VC,每个区对应一条虚通道,所有虚通道以时分复用方式共享该物理通道.由此可以是其它数据包绕过阻塞的数据包,提高物理通道利用率和网络吞吐率。 The basic idea is disposed on a plurality of buffers each physical channel, i.e. virtual channel VC, each zone corresponding to a virtual channel, time division multiplexed manner all the virtual channels share the physical channel thereby bypassing other data packets may be packet blocked, enhance the physical network channel utilization and throughput. 同时虚通道的引入可以较容易地实现自适应无死锁算法。 While introducing an adaptive virtual channel may be deadlock free algorithms implemented relatively easily. 属于同一个数据包的多个微片独占同一个VC,这就需要头微片从多个虚通道通过高效的仲裁机制申请到可用的VC。 A plurality of flits belonging to the same packet exclusive same VC, which requires the application to the VC header flit from a plurality of virtual channels available through efficient arbitration mechanism.

[0028] 交叉开关的申请仲裁器系指由于交叉开关作为多个端口直接的直连通路,需要n: 1的仲裁器从多个并发的端口请求中选择高优先级的微片先行传输。 [0028] Application crossbar arbiter means as a plurality of ports due to the crossbar direct passage directly connected, requires n: 1 arbiter selecting high priority is transmitted from the first plurality of flits ports of concurrent requests.

[0029] 本发明的技术方案是在通过应用的不同等级可靠性信息,通信数据微片的可靠性相关的感知信息指导路由器中VCA和SA仲裁优先级的配置。 [0029] aspect of the present invention is a VCA and SA are arranged in the arbitration priority guidance sensing information routers reliability reliability information different levels of application, the communication data relating to the flits.

[0030] 和传统的无可靠感知的仲裁方法不同,在本发明中,仲裁优先级的基于可靠性信息配置策略考虑的主要因素有: [0030] The conventional method of arbitration and no reliable perception, in the present invention, the arbitration priority factors are reliability information based on the configuration policies to consider:

[0031] (a)微片仲裁的历史信息:基于优先级的仲裁必须解决的一个关键问题就是“饿死”,即防止优先级低的微片的请求永远无法得到响应。 [0031] (a) historical information flits arbitration: based on a key issue that must be addressed arbitration priority is "starving to death" that prevent low-priority request flits never get a response. 尤其是在基于虫洞交换机制的网络传输中,“饿死”会导致尾随在优先级低微片之后的微片都被堵塞,引起网络局部停滞,进而影响整个网络通信质量。 Especially in the wormhole network transmission based on the switching mechanism, "starved" will lead to low priority after the trailing sheet flits are blocked, causing stagnant parts of the network, thereby affecting the overall quality of network communications. 针对此问题,本发明拟记录微片仲裁的历史信息,有效避免饿死。 To solve this problem, the present invention is intended to record history information flits arbitration, effectively avoid starvation. 同时,为了减少实现开销,对于记录历史信息的计数器进行循环取模操作,譬如2,4等经验值。 Meanwhile, in order to realize cost reduction, the recording history information for the counter is a modulo operation cycle, such as 2,4 experience value and the like. 记录历史信息的计数器的值随着仲裁成功或失败动态改变,具体为当仲裁成功则计数器值重置为零,否则进行加1取模操作。 The counter value of the recording history information with the success or failure of the arbitration dynamically changed, particularly when the arbitration is successful, the counter value is reset to zero, or are incremented modulo operation.

[0032] (b)应用的可靠性需求:基于一个应用中传输数据的重要性不同,它对可靠性的需求等级也有差异。 [0032] (b) the reliability of the application requirements: the importance of an application based on different data transfer, it needs the reliability level also varies. 因此,可靠性的需求等级表现为不同等级。 Therefore, the reliability level of demand for different performance levels. 举例来说,可把网络传输的数据粗略分为两级,即,高可靠性需求和一般可靠性需求。 For example, the data may be transmitted over the network are roughly classified into two, i.e., demand for high reliability and general reliability requirements. 即只需η比特的标志位来记录可靠性的需求等级。 I.e., bit flag only η reliability needs to record level. 同时,鉴于可靠性需求的生存周期是从进入网络开始到离开网络为止,且可靠性需求的标志位的值保持不变,所以由各微片携带此标志位。 At the same time, in view of the life cycle reliability requirements are beginning to enter the network from the network to the departure date, and the value of the flag reliability requirements remain unchanged, each microchip carry this flag.

[0033] 在本发明中,NoC中可靠性感知的仲裁机制如图2所示,通过上述两种可靠性感知仲裁的相关信息:应用的可靠性需求,微片仲裁的历史信息协同Round Robin的轮询仲裁进行优先级设置和选择。 [0033] In the present invention, the NoC arbitration mechanism perceived reliability of 2, perceptually relevant information by the two arbitration reliability: reliability requirements of the application, the history information flit synergistic Round Robin Arbitration of polling arbitration priority setting and selection.

[0034] 现详细描述根据本发明的片上互连网络的基于可靠性需求的仲裁方法。 [0034] The reliability requirements arbitration method based on a sheet according to the present invention will now be described in detail interconnection network.

[0035] 首先,为数据包各微片设置可靠性信息。 [0035] First, the reliability information is provided to each data packet flit.

[0036] 一实施例中,所述可靠性信息包括历史信息,所述历史信息记录仲裁成功或失败的动态改变。 [0036] In one embodiment of dynamic change, the reliability information includes history information, the history information recording the success or failure of the arbitration embodiment. 本实施例中,所述历史信息由计数器表示。 In this embodiment, the history information is represented by a counter. 当仲裁成功,则所述计数器的值重置为零,否则进行加1取模操作。 When the arbitration is successful, then the counter value is reset to zero, otherwise are incremented modulo operation. 为了减少实现开销,对于记录历史信息的计数器进行循环取模操作,譬如2,4等经验值。 In order to realize cost reduction, the recording history information for the counter is a modulo operation cycle, such as 2,4 experience value and the like. 记录历史信息的计数器的值随着仲裁成功或失败动态改变,具体为当仲裁成功则计数器值重置为零,否则进行加1取模操作。 The counter value of the recording history information with the success or failure of the arbitration dynamically changed, particularly when the arbitration is successful, the counter value is reset to zero, or are incremented modulo operation.

[0037] 另一实施例中,所述可靠性信息还包括可靠性需求信息,并且所述可靠性需求信息包括高可靠性需求和一般可靠性需求。 [0037] In another embodiment, the demand information further comprises a reliability information reliability, and the reliability of the information requirements include high reliability requirements and reliability requirements in general. 如前所述,基于一个应用中传输数据的重要性不同,它对可靠性的需求等级也有差异。 As described above, the importance of an application based on different data transfer, it needs the reliability level also varies. 因此,可靠性的需求等级表现为不同等级。 Therefore, the reliability level of demand for different performance levels.

[0038] 本实施例中,例如可把网络传输的数据粗略分为两级,S卩,高可靠性需求和一般可靠性需求。 [0038] In this embodiment, for example, the data transmission network are roughly classified into two, S Jie, high reliability requirements and reliability requirements in general. 由此,即只需η比特的标志位来记录可靠性的需求等级。 Thereby, i.e. η flag bits simply to record level needs reliability. 本实施例中,所述可靠性需求信息用1比特的标志位来记录,并且所述标志位设在所述数据包的微片中。 In this embodiment, the demand information with a reliability flag 1 bit is recorded, and the flit flag bit provided in the data packet. 由于整个通信带宽是一样的,所以无论是头,数据还是尾微片都添加此标志位,因此,所述标志为放在每个微片中。 Since the entire communication bandwidth is the same, so it is the head, or tail flit of data are added to this flag, so the flag is not on every flit.

[0039] 接着,根据所述可靠性信息对所述数据包微片的仲裁请求进行响应。 [0039] Next, according to the reliability information the arbitration request packet flit respond.

[0040] 由多个微片组成的各个数据包以微片为基本单位都需要仲裁,即VCA和SA保证通信正常进行。 [0040] Each packet flits composed of a plurality of micro-chip units are required as the basic arbitration, i.e. VCA SA and ensure normal communication. 每当进行VCA和SA时,需要发出仲裁请求。 Whenever conduct and VCA SA, need to issue a request for arbitration. 仲裁请求以对应的1位硬件信号线存在,独立于数据包。 An arbitration request signal line corresponding to the hardware is present, independent of the packet.

[0041] 在多个虚通道的头微片进行VCA仲裁时,同时发生请求的微片构成一个逻辑上的请求队列。 When [0041] for arbitration at the header flit VCA plurality of virtual channels, simultaneously request flits constituting a logical request queue. 此外,在多个端口的各种类型微片进行SA仲裁,也形成请求队列。 Further, a plurality of ports for arbitrating SA various types microchip, the request queue is also formed.

[0042] 当仲裁成功时,微片对应的历史信息计数器重置为0。 [0042] When a successful arbitration, the history information corresponding flit counter is reset to 0. 反之,仲裁失败队列对应的历史信息计数器则进行加1取模操作。 Conversely, failure of the arbitration queue counters corresponding to the history information are incremented modulo operation. 在多个端口的各种类型微片进行SA仲裁时操作和上述操作类似。 And the operation similar to the operation when SA flit arbitrates multiple ports of various types.

[0043] 具体地,对于一个特定的仲裁请求来说:仲裁成功则计数器值重置为零,否则就是仲裁失败则进行加1取模操作。 [0043] In particular, for a particular request for arbitration: the arbitration counter value is reset to zero if successful, otherwise the arbitration fails are incremented modulo operation. 而对于所有请求来说,就是仲裁成功计数器值重置为零,仲裁失败队列中所有对应的历史信息计数器均进行1取模操作。 For all requests, successful arbitration counter value is reset to zero, arbitration failure history information corresponding to all queue counters are modulo 1 operation performed.

[0044] —些实施例中,若多个所述微片的所述历史信息相同,则根据所述可靠性需求信息对所述历史信息相同的多个微片进行仲裁。 [0044] - In some embodiments, if a plurality of pieces of the same micro-history information, arbitration is performed in accordance with the same reliability requirement information for the plurality of history information flits. 由于可能有多个微片的历史信息相同,因此,在此种情况下,需要根据历史信息相同的多个微片的可靠性需求信息进行仲裁。 Since there may be a plurality of flits same history information, and therefore, in this case, the need for arbitration in accordance with the same reliability requirement information history information of a plurality of flits. 本实施例中,对可靠性需求信息为高可靠性需求的微片进行优先响应。 In this embodiment, the reliability of the demand information in response to the priority flits high reliability requirements.

[0045] 另一些实施例中,若多个所述微片的所述可靠性需求相同,则使用轮询仲裁进行优先级设置和选择。 [0045] In other embodiments, a plurality of the same if the flit reliability requirements, the polling arbitration priority setting and selection. 由于可能有多个微片的可靠性需求信息相同,因此,在此种情况下,则使用业界常用的一般仲裁办法进行仲裁。 Since there may be multiple flits reliability requirements of the same information, and therefore, in such cases, the industry commonly used general arbitration approach is used for arbitration. 本实施例中,采用经典的Round Robin进行仲裁。 In this embodiment, classic Round Robin arbitration. 现参考图2描述根据本发明的仲裁方法的具体实现。 2 depicts a specific implementation of an arbitration method of the present invention Referring now to FIG.

[0046] 在步骤SlOl中,完成仲裁的初始化工作,保证开始新一轮的仲裁选择。 [0046] In step SlOl, the arbitration is completed initialization, start a new round of arbitration to ensure that selection.

[0047] 在步骤S102中,读取请求队列中M个请求信息,以便进行仲裁处理。 [0047] In step S102, the M request queue information reading request for arbitration.

[0048] 在步骤S103中,对读取的请求信息进行比较,按照历史信息的等级进行排序,选择出历史信息等级最高的一个,如果选择成功,此次仲裁失败队列中的其他请求对应的历史信息的计数器进行加1取模操作,并跳转到步骤S106操作。 [0048] In step S103, a request for information read by comparing sorted by level of historical information, select the highest historical level of information, if you choose the success that the other historic corresponding to the request queue arbitration loss counter information are incremented modulo operation, and the operation jumps to step S106.

[0049] 在步骤S104中,如果N个请求历史信息等级相同,则再按照应用可靠需求的标志位进行排序,如果是高可靠性需求,则优先级高,如果选择成功,此次仲裁失败队列中的其他请求对应的历史信息的计数器进行加1取模操作,则跳转到步骤S106操作。 [0049] In step S104, if N requests history information the same level, the flag bit and then sorted according to the application requirements of reliable, if the demand for high reliability, the high priority, if the selection is successful, the failure of the arbitration queue the additional information corresponding to the request history counters are incremented modulo operation, the operation jumps to step S106.

[0050] 在步骤S105中,如果L个请求可靠性需求等级相同,则采用经典的Round Robin进行仲裁,如果选择成功,此次仲裁失败队列中的其他请求对应的历史信息的计数器进行加1取模操作,则跳转到步骤S107操作。 [0050] In step S105, if the request is the same as the L level reliability requirements, the classic Round Robin arbitration, if the selection is successful, the other counters request arbitration failure history information corresponding to the queue by incrementing the fetch mode operation, the operation jumps to step S107.

[0051] 在步骤106中,对选中的请求进行响应和服务,并准备进入新一轮仲裁,跳回步骤S101。 [0051] In step 106, for the selected service request and response, and ready to enter a new round of arbitration, jumps back to step S101.

[0052] 本实施例中,M、N、L俱为自然数。 [0052] In this embodiment, M, N, L all is a natural number.

[0053] 应用本发明可以充分优化片上互连网络中VCA和SA的仲裁机制,能够保证高可靠性需求的应用优先传输,减少在网络中的通信延迟,减少传输过程中软错误发生概率。 [0053] Application of the present invention can fully optimize the on-chip interconnect network VCA SA and arbitration mechanisms, it is possible to ensure the transmission of high priority application reliability requirements, to reduce communication delay in the network, reduce the probability of a soft error occurs during transmission. 同时还采用仲裁历史信息解决饿死问题,保证整个可靠感知的仲裁方法的有效和高效,具有重要实用价值。 It also uses historical information to solve starvation problem arbitration, the arbitration process to ensure the overall effective and efficient and reliable perception has important practical value.

[0054] 因本技术领域的技术人员应理解,本发明可以以许多其他具体形式实现而不脱离本发明的精神或范围。 [0054] by those skilled in the art to be understood that the present invention may be practiced without departing from the spirit or scope of the invention in many other specific forms. 尽管业已描述了本发明的实施例,应理解本发明不应限制为这些实施例,本技术领域的技术人员可如所附权利要求书界定的本发明精神和范围之内作出变化和修改。 Although embodiments have been described embodiments of the present invention, it should be understood that the present invention should not be limited to these embodiments, those skilled in the art may be appended claims the spirit and scope of the invention as defined in the book variations and modifications are within.

7 7

Claims (7)

  1. 1. 一种片上互连网络的基于可靠性需求的仲裁方法,其特征在于,包括如下步骤:(1)为数据包的各微片设置可靠性信息;(2)根据所述可靠性信息对所述微片的仲裁请求进行响应。 1. The arbitration method based on reliability requirements, characterized in that the on-chip interconnection network, comprising the steps of: (1) each data packet flit reliability information is set; (2) according to the reliability information the arbitration response request flit.
  2. 2.如权利要求1所述的仲裁方法,其特征在于,所述可靠性信息包括历史信息,所述历史信息记录仲裁成功或失败的动态改变。 2. The arbitration method according to claim 1, wherein the reliability information includes history information, the history information recording arbitration success or failure of the dynamic change.
  3. 3.如权利要求2所述的仲裁方法,其特征在于,所述历史信息由计数器表示,其中当仲裁成功,则所述计数器的值重置为零,否则进行加1取模操作。 The arbitration method as claimed in claim 2, wherein the history information is represented by a counter, wherein when the arbitration is successful, then the counter value is reset to zero, otherwise are incremented modulo operation.
  4. 4.如权利要求1所述的仲裁方法,其特征在于,所述可靠性信息还包括可靠性需求信息,并且所述可靠性需求信息包括高可靠性需求和一般可靠性需求。 The arbitration method as claimed in claim 1, wherein said reliability information comprises further information reliability requirements, the demand information and reliability requirements include high reliability and general reliability requirements.
  5. 5.如权利要求4所述的仲裁方法,其特征在于,所述可靠性需求信息用η比特的标志位来记录,并且所述标志位设在所述微片中。 5. The arbitration method according to claim 4, wherein said information recording reliability requirements η flag bits and said flag bit provided in the flit.
  6. 6.如权利要求5所述的仲裁方法,其特征在于,若多个所述微片的所述历史信息相同,则根据所述可靠性需求对所述历史信息相同的多个微片进行仲裁。 6. The arbitration method of claim 5, wherein, if a plurality of the same information history flit, then the reliability requirements according to the same history information arbitrating a plurality of flits .
  7. 7.如权利要求6所述的仲裁方法,其特征在于,若多个所述微片的所述可靠性需求相同,则使用轮询仲裁进行优先级设置和选择。 7. The arbitration method according to claim 6, wherein, if the plurality of the same flit reliability requirements, the polling arbitration priority setting and selection.
CN 201110358841 2011-11-14 2011-11-14 Reliability demand-based arbitration method in network on chip CN102394829A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201110358841 CN102394829A (en) 2011-11-14 2011-11-14 Reliability demand-based arbitration method in network on chip

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201110358841 CN102394829A (en) 2011-11-14 2011-11-14 Reliability demand-based arbitration method in network on chip

Publications (1)

Publication Number Publication Date
CN102394829A true true CN102394829A (en) 2012-03-28

Family

ID=45862041

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201110358841 CN102394829A (en) 2011-11-14 2011-11-14 Reliability demand-based arbitration method in network on chip

Country Status (1)

Country Link
CN (1) CN102394829A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102710530A (en) * 2012-06-28 2012-10-03 黑龙江大学 Configurable network-on-chip fault tolerance method
CN105022717A (en) * 2015-06-04 2015-11-04 中国航空无线电电子研究所 Network on chip resource arbitration method and arbitration unit of additional request number priority
CN105306382A (en) * 2014-07-28 2016-02-03 华为技术有限公司 Data processing method of No-cache NOC and NOC electronic element

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1993001553A1 (en) * 1991-07-08 1993-01-21 Seiko Epson Corporation Microprocessor architecture capable of supporting multiple heterogeneous processors
US6880028B2 (en) * 2002-03-18 2005-04-12 Sun Microsystems, Inc Dynamic request priority arbitration
CN101593159A (en) * 2008-05-30 2009-12-02 英特尔公司 Using criticality information to route cache coherency communications
CN101989942A (en) * 2009-08-07 2011-03-23 无锡江南计算技术研究所 Arbitration control method, communication method, arbitrator and communication system

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1993001553A1 (en) * 1991-07-08 1993-01-21 Seiko Epson Corporation Microprocessor architecture capable of supporting multiple heterogeneous processors
US6880028B2 (en) * 2002-03-18 2005-04-12 Sun Microsystems, Inc Dynamic request priority arbitration
CN101593159A (en) * 2008-05-30 2009-12-02 英特尔公司 Using criticality information to route cache coherency communications
CN101989942A (en) * 2009-08-07 2011-03-23 无锡江南计算技术研究所 Arbitration control method, communication method, arbitrator and communication system

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102710530A (en) * 2012-06-28 2012-10-03 黑龙江大学 Configurable network-on-chip fault tolerance method
CN102710530B (en) * 2012-06-28 2015-01-14 黑龙江大学 Configurable network-on-chip fault tolerance method
CN105306382A (en) * 2014-07-28 2016-02-03 华为技术有限公司 Data processing method of No-cache NOC and NOC electronic element
US10084710B2 (en) 2014-07-28 2018-09-25 Huawei Technologies Co., Ltd. Data processing method of NOC without buffer and NOC electronic element
CN105022717A (en) * 2015-06-04 2015-11-04 中国航空无线电电子研究所 Network on chip resource arbitration method and arbitration unit of additional request number priority

Similar Documents

Publication Publication Date Title
US7277429B2 (en) Cell-based switch fabric with distributed scheduling
Chien et al. Planar-adaptive routing: Low-cost adaptive networks for multiprocessors
US8050256B1 (en) Configuring routing in mesh networks
Kumar et al. Token flow control
US20110103391A1 (en) System and method for high-performance, low-power data center interconnect fabric
US20020181455A1 (en) Cell-based switch fabric with inter-cell control for regulating packet flow
US20020118692A1 (en) Ensuring proper packet ordering in a cut-through and early-forwarding network switch
US20090122703A1 (en) Electronic Device and Method for Flow Control
US20060165112A1 (en) Multi-stage packet switching system with alternate traffic routing
US20080285562A1 (en) Flexible routing tables for a high-radix router
US20090141622A1 (en) Pinning and protection on link aggregation groups
US20140115298A1 (en) ASYMMETRIC MESH NoC TOPOLOGIES
US20050259574A1 (en) Method and apparatus for implementing scheduling algorithms in a network element
US7391786B1 (en) Centralized memory based packet switching system and method
US20130028090A1 (en) Router and chip circuit
US20060039370A1 (en) Low latency switch architecture for high-performance packet-switched networks
US20030097514A1 (en) Computer system I/O node
US20120099475A1 (en) NoC SYSTEM AND INPUT SWITCHING DEVICE
US20140098683A1 (en) Heterogeneous channel capacities in an interconnect
US20050195845A1 (en) Low cost implementation for a device utilizing look ahead congestion management
US20020062415A1 (en) Slotted memory access method
US20070208898A1 (en) Programmable bridge header structures
US6941252B2 (en) Striping data frames across parallel fibre channel links
Ahmad et al. Architecture of a dynamically reconfigurable NoC for adaptive reconfigurable MPSoC
US20050018665A1 (en) Software configurable cluster-based router using stock personal computers as cluster nodes

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)