英文翻譯本文出自《ComputerNetwork》第四版Andrew

1、英文翻譯：本文出自 Computer Network 第四版 Andrew S.Tanenbaum 著Network Performance MeasurementWhen a network performs poorly, its users often complain to the folks running it, demanding improvements. To improve the performance, the operators must first determine exactly what is going on. To find out what is rea

2、lly happening, the operators must make measurements. In this section we will look at network performance measurements. The discussion below is based on the work of Mogul (1993). 矚慫潤厲釤瘞睞櫪廡賴賃軔。矚慫潤厲釤瘞睞櫪廡賴賃。The basic loop used to improve network performance contains the following steps: 聞創(chuàng)溝燴鐺險(xiǎn)愛氌 譴凈禍測(cè)。聞創(chuàng)

3、溝燴鐺險(xiǎn)愛氌譴凈禍。Measure the relevant network parameters and performance. 殘騖樓諍錈瀨濟(jì)溆塹籟婭騍。 殘騖樓諍錈瀨濟(jì)溆 塹籟婭。Try to understand what is going on.Change one parameter.These steps are repeated until the performance is good enough or it is clear that the last drop of improveme nt has bee n squeezed out釅錒極額閉鎮(zhèn)檜豬訣錐顧葒。釅錒極

4、額閉鎮(zhèn)檜豬訣錐顧。Measurements can be made in many ways and at many locations (both physically and in the protocol stack). The most basic kind of measurement is to start a timer when beginning some activity and see how long that activity takes. For example, knowing how long it takes for a TPDU to be acknowle

5、dged is a key measurement. Other measurements are made with counters that record how often some event has happened (e.g., number of lost TPDUS). Finally, one is often interested in knowing the amount of something, such as the number of bytes processed in a certain time interval. 彈貿(mào)攝爾霽斃攬磚鹵廡詒爾。彈貿(mào)攝爾霽斃攬

6、磚鹵廡詒。Measuring network performance and parameters has many potential pitfalls. Below we list a few of them. Any systematic attempt to measure network performance should be careful to avoid these. 謀蕎摶篋飆鐸懟類蔣薔點(diǎn)鉍。謀蕎摶篋飆鐸懟類蔣薔點(diǎn)。Make Sure That the Sample Size Is Large EnoughDo not measure the time to send o

7、ne TPDU, but repeat the measurement, say, one million times and take the average. Having a large sample will reduce the uncertainty in the measured mean and standard deviation. This uncertainty can be computed using standard statistical formulas. 廈礴懇蹣駢 時(shí)盡繼價(jià)騷巹癩。廈礴懇蹣駢時(shí)盡繼價(jià)騷巹。Make Sure That the Samples

8、Are RepresentativeIdeally, the whole sequence of one million measurements should be repeated at different times of the day and the week to see the effect of different system loads on the measured quantity. Measurements of congestion, for example, are of little use if they are made at a moment when t

9、here is no congestion. Sometimes the results may be counterintuitive at first, such as heavy congestion at 10, 11, 1, and 2 o'clock, but no congestion at noon (when all the users are away at lun ch).煢楨廣鰳鯡選塊網(wǎng)羈淚鍍齊。煢楨廣鰳鯡選塊網(wǎng)羈淚鍍。Be Careful When Using a Coarse-Grained ClockComputer clocks work by incr

10、ementing some counter at regular intervals. For example, a millisecond timer adds 1 to a counter every 1 msec. Using such a timer to measure an event that takes less than 1 msec is possible, but requires some care. (Some computers have more accurate clocks, of course.) 鵝婭盡損鵪慘歷蘢鴛賴縈詰。鵝婭盡損鵪慘歷蘢鴛賴縈。To me

11、asure the time to send a TPDU, for example, the system clock (say, in milliseconds) should be read out when the transport layer code is entered and again when it is exited. If the true TPDUsend time is 300 sec, the differenee between the two readings will be either 0 or 1, both wrong. However, if th

12、e measurement is repeated one million times and the total of all measurements added up and divided by one millio n, the mea n time will be accurate to better tha n 1sec.籟叢媽羥卩為贍僨蟶練淨(jìng)櫧撻?；[叢媽羥為贍僨蟶練淨(jìng)櫧。Be Sure That Nothing Unexpected Is Going On during Your Tests 預(yù)頌圣鉉儐歲齦訝驊糴買闥。預(yù)頌圣鉉 儐歲齦訝驊糴買。Making measuremen

13、ts on a university system the day some major lab project has to be turned in may give different results than if made the next day. Likewise, if some researcher has decided to run a video conference over your network during your tests, you may get a biased result. It is best to run tests on an idle s

14、ystem and create the entire workload yourself. Even this approach has pitfalls though. While you might think nobody will be using the network at 3 A.M., that might be precisely when the automatic backup program begins copying all the disks to tape. Furthermore, there might be heavy traffic for your

15、wonderful World Wide Web pages from distant time zones. 滲 釤嗆儼勻諤鱉調(diào)硯錦鋇絨。滲釤嗆儼勻諤鱉調(diào)硯錦鋇。Caching Can Wreak Havoc with MeasurementsThe obvious way to measure file transfer times is to open a large file, read the whole thing, close it, and see how long it takes. Then repeat the measurement many more times to

16、 get a good average. The trouble is, the system may cache the file, so only the first measurement actually involves network traffic. The rest are just reads from the local cache. The results from such a measurement are essentially worthless (unless you want to measure cache performance). 鐃誅臥瀉噦圣騁貺頂廡縫

17、勵(lì)。 鐃誅臥瀉噦圣騁貺頂廡縫。Often you can get around caching by simply overflowing the cache. For example, if the cache is 10 MB, the test loop could open, read, and close two 10-MB files on each pass, in an attempt to force the cache hit rate to 0. Still, caution is advised unless you are absolutely sure you un

18、derstand the caching algorithm. 擁締鳳襪備訊顎輪爛薔報(bào)贏。擁締鳳襪備訊顎輪爛薔報(bào)。Buffering can have a similar effect. One popular TCP/IP performance utility program has been known to report that UDP can achieve a performance substantially higher than the physical line allows. How does this occur? A call to UDP normally ret

19、urns control as soon as the message has been accepted by the kernel and added to the transmission queue. If there is sufficient buffer space, timing 1000 UDP calls does not mean that all the data have been sent. Most of them may still be in the kernel, but the performance utility thinks they have al

20、l been transmitted. 贓熱俁閫歲匱閶鄴鎵騷鯛 漢。贓熱俁閫歲匱閶鄴鎵騷鯛。Understand What You Are MeasuringWhen you measure the time to read a remote file, your measurements depend on the network, the operating systems on both the client and server, the particular hardware interface boards used, their drivers, and other factor

21、s. If the measurements are done carefully, you will ultimately discover the file transfer time for the configuration you are using. If your goal is to tune this particular configuration, these measurements are fine. 壇摶鄉(xiāng)囂懺蔞鍥鈴氈淚躋馱。壇摶鄉(xiāng)囂懺蔞鍥鈴氈淚 躋。However, if you are making similar measurements on three d

22、ifferent systems in order to choose which network interface board to buy, your results could be thrown off completely by the fact that one of the network drivers is truly awful and is only getting 10 percent of the performance of the board.蠟變黲癟報(bào)倀鉉錨鈰贅籜葦。蠟變黲癟報(bào)倀鉉錨鈰贅籜。網(wǎng)絡(luò)性能的測(cè)量當(dāng)一個(gè)網(wǎng)絡(luò)的運(yùn)行效果很差的時(shí)候， 它的用戶通常會(huì)向網(wǎng)絡(luò)運(yùn)

23、行商抱怨并要求提高網(wǎng)絡(luò)的質(zhì) 量。為了改善網(wǎng)絡(luò)的性能， 網(wǎng)絡(luò)操作人員首先必須確定發(fā)生了什么問題。 為了找出真正的問 題所在，操作人員必須進(jìn)行測(cè)量工作。在這一小節(jié)中，我們來看一看網(wǎng)絡(luò)性能的測(cè)量問題。 下面的討論以 Mogul（1993） 的工作為基礎(chǔ)。 買鯛鴯譖曇膚遙閆擷凄屆嬌。買鯛鴯譖曇膚遙閆擷凄屆。 用來改善網(wǎng)絡(luò)性能的基本循環(huán)過程包括以下步驟： 測(cè)量有關(guān)的網(wǎng)絡(luò)參數(shù)和性能。試圖理解當(dāng)前的網(wǎng)絡(luò)狀況。 改變一個(gè)參數(shù)。這些步驟不斷重復(fù)， 直到網(wǎng)絡(luò)的性能已經(jīng)足夠好， 或者改善性能的全部空間都已經(jīng)被發(fā)掘出 來了。測(cè)量工作可以有許多做法， 也可以在許多地點(diǎn)或場(chǎng)所進(jìn)行 （既指物理位置， 也指協(xié)議棧中的 位置）

24、。最基本的一種測(cè)量手段是：在開始某一個(gè)動(dòng)作的時(shí)候啟動(dòng)一個(gè)定時(shí)器，然后確定該 需要多長時(shí)間。例如，知道一個(gè) TPDU 需要多長時(shí)間才被確認(rèn)是一個(gè)很關(guān)鍵的測(cè)量指標(biāo)。 其他有一些測(cè)量指標(biāo)可以通過計(jì)數(shù)器來完成，即記錄某種事件發(fā)生的次數(shù)，比如丟失的 TPDU 的數(shù)量。最后，人們通常對(duì)于某些事物的數(shù)量比較感興趣，比如在特定的時(shí)間間隔內(nèi) 所處理的字節(jié)數(shù)。 綾鏑鯛駕櫬鶘蹤韋轔糴飆鈧。綾鏑鯛駕櫬鶘蹤韋轔糴飆。 測(cè)量網(wǎng)絡(luò)的性能和參數(shù)有許多潛在的陷阱。 以下我們列出其中一部分。 任何一種系統(tǒng)化的網(wǎng) 絡(luò)性能測(cè)量手段都應(yīng)該小心地避免這些陷阱。 驅(qū)躓髏彥浹綏譎飴憂錦諑瓊。驅(qū)躓髏彥浹綏譎飴憂錦諑。 確保樣本空間足夠大不要測(cè)

25、量發(fā)送一個(gè) TPDU 的時(shí)間，而是重復(fù)也測(cè)量。比如說測(cè)量 1 百萬次，然后再取平均。 采用大量的樣本將可以減小所測(cè)量的均值和標(biāo)準(zhǔn)方差中的不確定性。 這種不確定性可以利用 標(biāo)準(zhǔn)的統(tǒng)計(jì)公式來計(jì)算。 貓蠆驢繪燈鮒誅髏貺廡獻(xiàn)鵬。貓蠆驢繪燈鮒誅髏貺廡獻(xiàn)。確保樣本具有代表性理想情況下， 這 1 百萬次測(cè)量的完整序列應(yīng)該在一天或者一周的不同時(shí)刻進(jìn)行重復(fù)， 從而可 以看到不同的系統(tǒng)負(fù)載對(duì)于所測(cè)量指標(biāo)的影響。 例如， 對(duì)于擁塞的測(cè)量， 如果僅僅在沒有擁 塞的那一時(shí)刻來測(cè)量擁塞， 則這樣的測(cè)量和結(jié)果并沒有用。 有時(shí)候測(cè)量結(jié)果初看起來可能不 符合直覺，比如在 10，11，1 和 2 點(diǎn)鐘網(wǎng)絡(luò)嚴(yán)重?fù)砣侵形鐣r(shí)候沒

26、有擁塞（所用的用戶 都去吃午飯了） 。 鍬籟饗逕瑣筆襖鷗婭薔嗚訝。鍬籟饗逕瑣筆襖鷗婭薔嗚。當(dāng)使用粗粒度時(shí)鐘的時(shí)候一定要謹(jǐn)慎計(jì)算機(jī)時(shí)鐘的工作原理是， 每隔固定的時(shí)間間隔就遞增某一個(gè)計(jì)數(shù)器，例如， 一個(gè)毫秒定時(shí)器每隔 1ms 就讓一個(gè)計(jì)數(shù)器加 1。使用這樣的定時(shí)器來測(cè)量一個(gè)持續(xù)時(shí)間小于1ms 的事件是有可能的，但要非常小心。 （當(dāng)然，有些計(jì)算機(jī)還有更加精確的時(shí)鐘。 ） 構(gòu)氽頑黌碩飩薺齦話騖 門戲。構(gòu)氽頑黌碩飩薺齦話騖門。例如，為了測(cè)量出發(fā)送一個(gè) TPDU 所需要的時(shí)間，當(dāng)進(jìn)入傳輸層代碼時(shí)以及離開傳輸層代碼時(shí)，應(yīng)該將系統(tǒng)時(shí)鐘（比如說以毫秒為單位） 讀出來。如果TPDU真正的發(fā)送時(shí)間是 300陰, 則

27、兩次讀取的時(shí)間之差要么是0,要么是1,這兩個(gè)結(jié)果都是錯(cuò)誤的。 然而，如果重復(fù)測(cè)量1百萬次，則所有測(cè)量的總和累加起來，再除以1百萬，則平均時(shí)間比1Q還要精確得多。輒嶧陽檉籪癤網(wǎng)儂號(hào)澩蠐鑭。輒嶧陽檉籪癤網(wǎng)儂號(hào)澩蠐。確保在測(cè)試過程中不會(huì)發(fā)生不可預(yù)知的事情 在一個(gè)大學(xué)的網(wǎng)絡(luò)系統(tǒng)進(jìn)行測(cè)量有可能發(fā)生這樣的情況： 有一天, 當(dāng)一個(gè)大型的實(shí)驗(yàn)項(xiàng)目在 運(yùn)行的時(shí)候你測(cè)量的結(jié)果跟第二天測(cè)量出來的結(jié)果可能會(huì)有所不同。同樣地, 如果有的研究人員決定在你們的網(wǎng)絡(luò)上運(yùn)行一個(gè)視頻會(huì)議, 而在這個(gè)時(shí)候你正好在測(cè)量, 那么你得到的結(jié) 果可能會(huì)偏差。 你最好在一個(gè)空閑的系統(tǒng)上運(yùn)行測(cè)試過程, 并且根據(jù)需要自己來創(chuàng)建所有的 工作負(fù)載。 不過這種做法也有缺陷。 你可能認(rèn)為在凌晨 3 點(diǎn)鐘的時(shí)候不會(huì)有人使用網(wǎng)絡(luò), 但是，當(dāng)自動(dòng)備份程序這時(shí)候開始將所有的磁盤數(shù)據(jù)復(fù)制到磁帶上的時(shí)候， 你的想法就不再正 確了。而且，此時(shí)其他時(shí)區(qū)的用戶可能會(huì)訪問你精美的 WWW, 從而也導(dǎo)致繁重的流量。 堯側(cè) 閆繭絳闕絢勵(lì)蜆贅瀝紕。堯側(cè)閆繭絳闕絢勵(lì)蜆贅瀝。緩存機(jī)制可能會(huì)破壞測(cè)量的正確性為了測(cè)量文件傳輸時(shí)間， 最顯然的方法是打開