網路的使用愈來愈普及，相對的也發展出許多新的服務及應用來
因應使用者的須求。為了加快這些服務的處理速度，網路設備廠商就

會將這些網路服務製成硬體ASIC，提昇處理效能。但開發ASIC所需的

時間冗長費時，往往趕不上新的服務所發展的速度；另一方面，製成

ASIC後，若有問題亦不能輕易修改，因此所需的人力物資成本非常高

，為了解決原先網路設備開發上的缺點，於是就有了網路處理器

(Network Processor) 的概念；其主要構想是設計一顆專門用來處理

網路資訊處理器，可以利用它來設計相關處理網路服務的程式。好處

是它內部即有處理網路資訊的指令，可減少開發產品的時間；另外，

它可依使用者所須，開發不同的應用，並且可以隨時修正更改，減少

設計錯誤造成的風險，增加網路設備設計上的彈性，因此其為將來網

路處理最佳解決方案。

但網路處理器的概念目前尚未成熟。就我們所選擇的網路處理器

而言，由於其硬體架構的關係，會有與縱橫式(Crossbar)交換器類似

的前端資料阻塞(Head of line blocking) 問題，因此我們設計虛擬

輸入佇列(VIQ) ，讓輸入封包標頭先轉送到記憶體，使新的封包能夠

及時被處理，解決輸入端的瓶頸。另外若輸出佇列只有一個，將不能

及時處理大量連續的輸出封包，因此設計了動態輸出佇列(DOQ) ，使

連續輸出資料不會受影響，如此，一方面可補強原先輸入端設計架構

的不足，另一方面讓原先輸出端的架構，能發揮最好的使用性。

動態輸出佇列的情況下，所能提昇的效能及結果分析，讓網路處

理器發揮最佳的效能。

Because of the more and more popularity of Internet,
more and more new services and applications are created to

meet the user's requirements. In order to speed up the ser-

vice processing, many of the network manufactures like to

implement such network services into Hardware ASIC to enhance

the performance. But it takes a very long time to design an

ASIC so that the speed of development time could not meet the

requirement of the demands of new services, and cost is still

an important issue as well. That's why we need Network Pro-

cessor, a brand new concept to solve these probelms. The main

idea of NP is to design a programmable processor that is able

to process the network information case by case according to

the micro code inside. By the help of micro code,the enter-

prises could save the development time. Because NP is programm-

able and even in case modification is necessary, all you have

to do is downloading a new image code into NP instead of

making another new hardware instance. The whole hardware

architecture could remain unchanged. It goes without saying

that user could develope any kind of network applications on

their own demand under the NP architecture. Still the same,

in most case the only thing needed to do is putting a new

image code into the NP.

Although there are several companies making good NPs, but

the prefect NP is still not available and the system perfor-

mance will be bounded by NP design. Let's take the NP we

choosed as an example, it will have the same problem named

'head of line blocking' with the common Crossbar Switch because

of its hardware design. In order to solve that problem, we've

designed a Virtual Input Queue(VIQ) as a solution. Another pro-

blem is that there is only one output queue inside the NP, and

it could be happened to lost packets in burst condition. So

that we design the Dynamic Output Queue (DOQ) to solve this

problem.

Finally we designed several test cases for analyzing accor-

ding to the above architecture, and we've proved that the NP

could raise its performance by the help of VIQ and DOQ.

[1]. 我國網路用戶數成長情形, "http://www.find.org.tw/img_focus/focus20010306_1.gif"
[2]. 目前台灣人口統計, http://www.teputc.org.tw/env_news/199906/88061901.htm
[3]. Intel Corporation web site, "http://www.intel.com/"
[4]. IBM Corporation web site, "http://www.ibm.com/"
[5]. MMC Corporation web site, "http://www.mmc.com/"
[6]. VITESSE Corporation web site, http://www.vitesse.com/
[7]. IQ2000 Network Processor, http://www.vitesse.com/products/categories.cfm?family_id=5&category_id=16
[8]. Massoud R. Hasemi, Alberto Leon-Garcia, "A Multicast Single-Queue Switch with a Novel Copy Mechanism", IEEE INFO- COM'98, San Francisco California, 1998
[9]. Cheng-Shang Chang, Wen-Jyh Chen, Hsiang-Yi Huang, " Birk-hoff-von Neumann Input Buffered Crossbar Switches ", IEEE INFOCOM'00, Tel Aviv, Israel, 2000, pp. 1614 -- 1623
[10]. Shang-Tse Chuang, Ashish Goel, Nick McKeown, Balaji Prab-hakar, "Matching Output Queuing with a Combined Input Out-put Queued Switch", IEEE INFOCOM'99, New York, March 1999, pp. 1169-- 1178
[11]. Mark W. Goudreau, Stavros G. Kolliopoulos, Satish B. Rao, "Scheduling Algorithms for Input-Queued Switches: Randomi- zed Techniques and Experimental Evaluation", IEEE INFO- COM'01, Alaska USA, April 2001
[12]. D. N. Sepanos, P. I. Antoniadis, "FIRM: A Class of Distributed Scheduling Algorithms for High-speed ATM Switches withMultiple Input Queues", IEEE INFOCOM'01, Alaska USA, April 2001
[13]. Adisak Mekkittikul, Nick McKeown, "A Practical Scheduling Algorithm to Achieve 100% Throughput in Input-Queue Swit- ches", IEEE INFOCOM'98, San Francisco California, 1998
[14]. Matthew Andrews, Sanjeev Khanna, Krishan Kumaran, "Integ- rated Scheduling of Unicast and Multicast Traffic in an Input- Queued Switch", IEEE INFOCOM'98, San Francisco Califor- nia, 1998
[15]. Shizhao Li, Nirwan Ansar, "Input-Queued Switching with QoS Guarantees", IEEE INFOCOM'98, San Francisco California, 1998
[16]. Matthew Andrew, Lisa Zhang, "Achieving Stability in Network of Input-Queued Switches", IEEE INFOCOM'01, Alaska USA, April 2001
[17]. M. Ajmone Marsan, E. Leonardi, M. Mellia, F. Neri, "On the Stability of Input-Buffer Cell Switches with Speed-up", IEEE INFOCOM'01, Alaska USA, April 2001
[18]. E. Leonardi, M. Mellia, F. Neri, M. Ajmone Marson, "Bounds on Average Delays and Queue Size Averages and Variances in Input-Queued Cell-Based Switches", IEEE INFOCOM'01, Ala- ska USA, April 2001