協力式Web代理伺服器之研究設計---子計畫三：協力式Web代理伺服器之協力式快取之研究設計(I)

全文

(1)

(2) Web

(3) (I) Design of Collaborative Caching for Collaborative Web Proxy Server NSC87-2213-E-009-015 86

(4) 8 1 87

(5) 7 31 . . !. are some problems in performance of proxy servers. World Wide WebWWW In this project, we propose a method 1993

(6)

(7) based on a mechanism, called object location !"# table (OLT), that can combine the resources $

(8) %&'()*+&,-./ of Web proxy servers and make information 012

(9) 345678 9:;< retrievals more efficient. In our experiments, = + > ? Web @ A ! " proxy it shows this method has better efficiency. server-BCDEFG=

(10) H Keyword: World Wide Web (WWW)pWeb I Web @A !"JKLMN<OP DocumentpWeb ServerpWeb Proxy Serverp ?M12Hierarchical Cache Arrangement p Global QRS9T0UV)W XY Resource Indexing (GRI)pObject Location Object Location TableOLTZ8

(11) ? Table (OLT) [\] Web @A !"^

(12) _à *+b?=

(13) Cc] Web @A ! (#)*+,",dLF?eFfg

(14) hij`k ylzBraunz{|}~[

(15) l mnhio8 WWW JInternet MF

(16) pWeb ()pWeb !"pWeb @A !"pqrs=+tup k-/_4. vs^wxpV)WXY 12- 12 ¡*+F¢`£ Abstract World Wide Web (WWW) has been the £¤

(17) _`¥¦£ £§¨

(18) Cc most popular Internet service, since 1993. !"#©£ £¤

(19) /()* Thus, the information growth over WWW +ª«&,-z ./012

(20) 345678 9 also cause more load on network and servers.. "#$%&'. :;< cache >? Web @A. In order to solve this problem, the common solution is to support a proxy server with cache functions. However, due to the fast growth of the Internet information, there. !". proxy server-¬< Web @A !" Harvest cache ® Squid ¯<°±qrs =+tu

(21) _4²³U ICP(Internet 1.

(22) Cache Protocol) ´ µ ¶ · ¸ < ¹ Web ()

(23) º»¼\] Web @A ! "^-½ HIqrs=+tu¾ <¿À12. Inter ne t I nt ernet. Proxy Server OLT. ÁÂÃµMÄÅ-Æ< Web @A !"Ç&

(24) Èo®É\L Web @A !"¯ÊËÌÍÃµ-. Cache. Pr oxy Ser ve r OLT. ÁÎÂ¾ cache miss ÏÐ&,Á Â_4 ICP ·¸DEÑ?{ÒÃÓ µ&,

(25) Ô¸1ÕÖ¹8×¤-. Browser Browser Ca che. QØRSÙÚÛ[\] Web @ A !",^

(26) _à*+ bÌ=

(27) !"#©§¾9×

(28) C c§¾<MÀ12-z. Pr oxy Ser ve r O LT. Cac he. Browser Browser. Cac he. Br ow se rr Browse. Cache. p_4V)WXY=+qrtu ¿ ¾Æ0_4V) WXY Web @A !" cache miss &

(29) ã+à»-< 4 Web @A !" P *+ 0 Web () D

(30) J P Q!"#$§ (1) ·¸ P Q!V)WXY

(31) V)W XY%<ö¡û Web ()a

(32) ó& ICP Zs 'A(2) · Web @A !" PX ú< Web () D

(33) ( P PX *+ D-½J %&

(34) È<\Lö÷0 Web @ A !")T ICP query(3) PX J*+,-./ D

(35) P ó{ D .0

(36) B12.34-%&4ë D JV)WXY56 P Ãµ§ 78 PX .G Web ()

(37) (§44 ë-. .#/0-1234 Ü. Harvest cache ® Squid qrs =+tuÝÞ

(38) hi9T0uo ßàÉ-hiáâã¤ä Sun Wu ® Chang-Chain Liao ª9Tå³Uv s^wxGlobal Resource Indexing

(39) æ ç GRI[5] Z8

(40) ºC9Thièé 56Zê-Jhiuo

(41) 0 Web @A !"ë cache ìíìîï0 table ð ð h iñ & ò³ E OLT (Object Location Table)-Ñ4 Web () URL

(42) óô OLT ·THI<õ\Lö÷ø Web @A !"ù0ú<û( )-ªü\Lö÷

(43) |dJ Web @ A !"ÃµýÃµ

(44) ZsþÃ. (4) 9Ã PX J*+/0{:&,ô; « Í%</<=>É%< D

(45) ( P ? ICP Zs 'AV)WXYÌ@EÙ®A*. µ squid parent ® sibling-_4 OLT =+qrtuªV)WXY® ICP query ? ê µ

(46) DE_4V)WXY. ²-0°±V)WXY Web @A ! "¯ÊËB<\Lö÷ Web @A ! "èéùO Web ()

(47) %&'C ÉiB-. B

(48) È Web @A !"·¸& ,

(49) ½V)WXYaõ<ø &

(50) Nqrs=+tu + à-. V)WXYÙÌ D

(51) Web @ 2.

(52) A. !"$TMAÙÌHI/. Web @A !"./34 Web ( )kHëª< Web ()nkC Bytes hit rate Web @A !"./34 Web ()ýê¤Sn®ëª< Web ()ýê¤Sn®Æof Web @A !""#p,§ q

(53) «rTO Web ()j"# 4ë&

(54) HI_4sI8 ²Z. E&,èéQ!"# Web ()

(55) cFÉi URL ÀY

(56) B1BÈ< \Lö÷ Web @A !"- ¡ÙÌ GHKI&,®¥¦

(57) ªJK LMA-N¡AÌOQMGþÙ Ì

(58) Ñ PQRS

(59) N*0S&M A® Web @A !"tuT

(60) tu ÙUÞ ôVW cache miss &

(61) 4 ICP È Web @A !"·¸ &,

(62) DE·¸V)WXYóôBX. LRU algorithmðð 32 0 Web () $T access time t¬J u0j 4ë

(63) &"#p,¨qEvhiwx^=+"#F. hit rate yz

(64) /0]^hi@i.j¤R®. YZù0 Web @A !"<hio Web ( )

(65) B 1 ó& É - C c Document Index kHôG

(66) Dû[ \ÕÖôG¤HI]^ª`[

(67) 0_4. j¤d log files ]M{]^-{ ]^]|x}²"#F

(68) @i ¹~ ¤pLfF

(69) Lf F 1/2p1/4p1/8p1/16-. _ Web @A !" P ®0_4_ Web @A !" Q

(70) Q V)WXY ` P ±à

(71) ô` Q P?a`b Íc-. Infini 1.331 6656 3328 1664 832.0 te G M M M M Counts 72.63 69.30 60.78 50.83 41.97 33.92 hit rate % % % % % % Bytes 54.49 51.25 44.29 36.71 30.08 23.99 hit rate % % % % % %. 5#6789:. Yp=+"#Fyzj¤R{. CC. QRdHIÙÚÛZ8

(72) e4. CSIE. Squid 1.1.x access log file

(73) ºQfg@ h-ª_4 access log files @ij¤ R Web @A !"®j¤d Web @A !"+`-HIk M]â Y®Ylª-. Counts hit rate Bytes hit rate. Yp=+"#Fyzj¤d{ *{]^[

(74) ` hit rate T ]KL Web @A !" hit rate `-D<¿À0pÆL*. 1997/11/ 1997/11 1997/11/ 1997/11/ 2 3 4 5 Count 902331 893996 1000654 924280 930315 Bytes 12.65G 12.22G 14.08G 13.64G 13.15G CC. Ypj¤R@A. 0 Web @A !" hit rate Q óGlphiof Web @A !"?¹8 Qf ã Web @A !"J#+M V-<Þ`

(75) Object bytes hit rate ¯T Object counts hit rate

(76) / J/LSýê_4T¤ýê `-. !" access log file. CSIE 1997/11/ 1997/11/ 1997/11/ 1997/11 2 3 4 /5 Count 9947 23144 38293 39909 27823 Bytes 211.0M 473.5M 708.3M 821.5M 553.6M. Ylpj¤d@A. Infini 550.0 275.0 137.5 68.75 34.38 te M M M M M 62.06 55.76 46.28 38.57 32.31 26.35 % % % % % % 53.90 49.47 40.88 32.28 24.75 18.01 % % % % % %. !" access log file. him^*² hit rate-Counts hit rate 3.

(77) hit rate. J]^l

(78) hio*0 Web @ A !"\

(79) ¾§¾` hit rate 9-L 8 j%L*0 Web @A !" access log files \

(80) +&, -1ó]^Z8 M

(81) =+ FÃµ'-. %. %. %. %. %. Bytes 61.95 60.87 58.01 55.53 53.63 51.44 hit rate % % % % % %. Y}p_4V)WXY[. ;#7<= M¡0]^

(82) ?§?V) WXY <¢à£»[. µ

(83) §G¤¥¦]§]4

(84) ?+. CC & Infinit 1.386 6931 3466 1733 866.4 CSIE e G M M M M Counts 72.72 69.19 60.76 51.11 42.40 34.56 hit rate % % % % % % Bytes 54.97 51.50 44.73 37.55 31.15 25.40 hit rate % % % % % %. Yp*0 Web @A. %. `Ì Web @A !" access logs ºQf-Cc§?Ñ<*§ Web @A !" log

(85) <¨§M©-%&ª s'?ÊËÍ«

(86) â¬Ì ]®

(87) V)WXY0S&,AÌ{¯ dM<0°±Þ

(88) . !"\. Yk®Y§

(89) Yj ¤R®j¤d/*0 Web @A !" \J

(90) .R Web @A !"_4_%<9 hit rate

(91) % '

(92) ÆBd Web @A !"_ 4_è§ -/ DEj¤R Web @ A !"_4_|k¡j¤d. Web @A !" log +`§²-lÞ

(93) hiQf?Mªb

(94) ³ Web @A !",ÏÐD´ â¬-

(95) BV)WXY?_·¸ ÕÖ¢¤

(96) ½? µ©¾< P-. Web @A !"]^ |x OLT P-hi9Ã /

(97) DEj¤R Web @A !"_4_|kR

(98) ªR Web @A !"¾Ìd Web @A !" V)WXY-

(99) J|F hit rate I

(100) d Web @A !"V)WXY <56R Web @A !"<ùO Web ()-N¡ Cache F®]^j¤ d{0 Y}[

(101) J§=+" #F¿

(102) ®Y[T

(103). >#?@%A [1] Michael Accetta. RFC 887: Resource location protocol, December 1983. [2] Guy T. Almes, Andrew P. Black, Edward D. Lazowska, and Jerr e D. Noe. The Eden system: A technic al revie w. IEEE Transactions on Softwa re Enginee ring, SE-11(1):43--59, January 1985. [3] Rafael Alonso and Matthe w Blaze. Long-te rm caching strategies for very large distributed file systems. Proceedings of the USENIX Summer Conference, pages 3--16, June 1991. [4] David R. Boggs. Internet broadcasting, January 1982. Ph.D. Thesis, available as Technical Report CSL-83-3, Xerox Palo Alto Research Center, October 1983. [5] Sun Wu and Chang-Chain Liao, “Virtual Proxy Servers for WWW and Intelligent Agents on the Internet ”, 30th Hawaii International Confer enc e on System Science, January 1997.. *² hit rate ¯<Íc- B/M ._4_ Web @A !"%< '

(104) ½ ._4_|k Web @A ! "?§Q

(105) §4 È_4_ ¿

(106) ` hit rate b9CSIE Infini 550.0 275.0 137.5 68.75 34.38 with GRI te M M M M M Counts 76.71 75.33 71.30 66.55 62.59 58.83 4.

(107)