NoC中VOQ路由器及鏈路容錯(cuò)技術(shù)研究
發(fā)布時(shí)間:2018-06-28 00:57
本文選題:片上網(wǎng)絡(luò) + 容錯(cuò)路由器 ; 參考:《合肥工業(yè)大學(xué)》2015年碩士論文
【摘要】:集成電路的不斷發(fā)展使得片上系統(tǒng)遇到了不可逾越的瓶頸問(wèn)題,片上網(wǎng)絡(luò)因其具有良好的可擴(kuò)展性、較高的通信效率和可靠性等優(yōu)勢(shì)成為了一種新型的互連架構(gòu),被研究者們廣泛研究。路由器是片上網(wǎng)絡(luò)中的關(guān)鍵部件,主要負(fù)責(zé)數(shù)據(jù)包的存儲(chǔ)轉(zhuǎn)發(fā),影響著網(wǎng)絡(luò)的整體性能,故對(duì)路由器的容錯(cuò)設(shè)計(jì)顯得尤為重要。NoC中數(shù)據(jù)包的傳輸需要經(jīng)過(guò)鏈路,且鏈路占據(jù)了芯片上的絕大部分面積,故對(duì)鏈路的容錯(cuò)尤為重要。論文主要工作如下:(1)闡述了集成電路的發(fā)展、片上網(wǎng)絡(luò)的研究背景。對(duì)常見(jiàn)的片上網(wǎng)絡(luò)拓?fù)浣Y(jié)構(gòu)及片上網(wǎng)絡(luò)中常見(jiàn)的容錯(cuò)方法進(jìn)行了介紹。(2)路由器中緩沖區(qū)所占的面積較大,且常見(jiàn)的緩沖區(qū)容錯(cuò)方案都可能會(huì)產(chǎn)生頭阻塞問(wèn)題,因此需要設(shè)計(jì)一種容錯(cuò)路由器架構(gòu),既可以解決頭阻塞問(wèn)題,又能實(shí)現(xiàn)緩沖區(qū)容錯(cuò)。本文提出一種相鄰虛通道循環(huán)共享的VOQ容錯(cuò)路由器設(shè)計(jì),可以有效地容忍虛通道故障,在網(wǎng)絡(luò)擁塞時(shí)能最大程度的保證緩沖區(qū)的利用率,且在路由器內(nèi)部虛通道均故障時(shí)依舊可以保證整個(gè)網(wǎng)絡(luò)的連通性。在基準(zhǔn)VOQ路由器的基礎(chǔ)上,該文通過(guò)增加多個(gè)多路選擇器和數(shù)據(jù)分配器來(lái)實(shí)現(xiàn)相鄰虛通道的共享,增加旁路總線來(lái)保證即使輸入端口內(nèi)部虛通道均故障,路由器依舊可用。同時(shí),該文設(shè)計(jì)了一種基于雙向隊(duì)列的FIFO控制器,通過(guò)兩套讀寫指針實(shí)現(xiàn)對(duì)借用虛通道數(shù)據(jù)包和非借用虛通道數(shù)據(jù)包的讀寫操作。(3)為解決鏈路的故障問(wèn)題,該文將鏈路故障進(jìn)行粒度劃分,并提出了一種自適應(yīng)的鏈路容錯(cuò)方法。當(dāng)鏈路出現(xiàn)粗粒度故障時(shí),重配置另一條鏈路的方向來(lái)傳輸數(shù)據(jù)包。當(dāng)鏈路出現(xiàn)細(xì)粒度故障時(shí),結(jié)合網(wǎng)絡(luò)通信狀態(tài),選擇重配置另一條鏈路的方向或利用自身剩余的完好位線來(lái)多次傳輸數(shù)據(jù)包。實(shí)驗(yàn)結(jié)果表明,該文方法較雙向鏈路重配置方案和單向鏈路多次傳輸?shù)娜蒎e(cuò)方案而言,延時(shí)最多減小27.2%和34.2%,吞吐率最多增大21.7%和35.3%。在鏈路出現(xiàn)不同情況的故障時(shí),該文方案有著明顯的優(yōu)勢(shì),保證了網(wǎng)絡(luò)的可靠性。
[Abstract]:The continuous development of integrated circuits makes the on-chip system meet an insurmountable bottleneck problem. Because of its good scalability, high communication efficiency and reliability, on-chip network has become a new type of interconnection architecture. Widely studied by researchers. Router is a key component in the on-chip network. It is mainly responsible for the storage and forwarding of data packets and affects the overall performance of the network. Therefore, the fault-tolerant design of routers is particularly important to the transmission of packets in .NoC through links. And the link occupies most of the chip area, so the fault-tolerance of the link is particularly important. The main work of this paper is as follows: (1) the development of integrated circuits and the research background of on-chip network are described. The common topology of on-chip network and the common fault-tolerant methods in the on-chip network are introduced. (2) the buffer occupies a large area in the router, and the common buffer fault-tolerant schemes may cause the head-blocking problem. Therefore, it is necessary to design a fault-tolerant router architecture, which can solve the header blocking problem and implement buffer fault tolerance. In this paper, a VOQ fault-tolerant router with adjacent virtual channel cycle sharing is proposed, which can effectively tolerate virtual channel failures and ensure the maximum buffer utilization rate when the network is congested. The connectivity of the whole network can still be guaranteed when the internal virtual channel of the router fails. Based on the reference VOQ router, this paper adds multiple multiplexers and data allocators to realize the sharing of adjacent virtual channels, and adds a bypass bus to ensure that the router is still available even if the internal virtual channels of the input ports fail. At the same time, this paper designs a FIFO controller based on bidirectional queue, which can read and write virtual channel data packet and non-borrowed virtual channel data packet through two sets of read and write pointers. (3) to solve the link fault problem, In this paper, link fault is partitioned by granularity, and an adaptive link fault tolerant method is proposed. When a coarse-grained link fails, reconfigure the direction of another link to transmit data packets. When there is a fine-grained failure of the link, combining with the network communication state, we choose to reconfigure the direction of another link or use its remaining intact bit lines to transmit data packets many times. The experimental results show that compared with the bidirectional link reconfiguration scheme and the one-way link multiple transmission scheme, the proposed method reduces the delay by 27.2% and 34.2%, and increases the throughput by 21.7% and 35.3%. The proposed scheme has obvious advantages and ensures the reliability of the network when the link failure occurs in different situations.
【學(xué)位授予單位】:合肥工業(yè)大學(xué)
【學(xué)位級(jí)別】:碩士
【學(xué)位授予年份】:2015
【分類號(hào)】:TN915.05;TN47
【相似文獻(xiàn)】
相關(guān)期刊論文 前3條
1 徐曉飛;;分布式調(diào)度算法在VOQ交換機(jī)中的應(yīng)用[J];計(jì)算機(jī)應(yīng)用;2006年08期
2 馬麗紅;蔡祥寶;;帶VOQ的輸入隊(duì)列交換網(wǎng)絡(luò)中的分組調(diào)度算法研究[J];光子技術(shù);2006年03期
3 ;[J];;年期
相關(guān)碩士學(xué)位論文 前1條
1 王悄;NoC中VOQ路由器及鏈路容錯(cuò)技術(shù)研究[D];合肥工業(yè)大學(xué);2015年
,本文編號(hào):2075992
本文鏈接:http://sikaile.net/kejilunwen/dianzigongchenglunwen/2075992.html
最近更新
教材專著