Share
Embed
FARIS: Fast and Memory-efficient URL Filter by Domain Specific Machine
ytakano
September 28, 2016
Transcript
17 slides
-
"E#MPDL1MVT`T4ZOUBY BOE3FHVMBS&YQSFTTJPOT "E#MPDL1MVT BOEPUIFSBQQMJDBUJPOTMJLF V#MPDL VTFTSFHVMBSFYQSFTTJPOTJOUFSOBMMZ TABLE I FILTER
SYNTAX OF ADBLOCK PLUS AND ITS REGULAR EXPRESSIONS AdBlock’s syntax regular expression * /.*/ | of the beginning of a line /ˆ/ | of the end of a line /$/ || of the beginning of a line /[\w\-]+\/+/ ˆ /[\x00-\x24\x26-\x2C\x2F\x3A-\x40\x5B-\x5E\x60\x7B-\x7F]|$/ a hash function to the pieces to minimize hash colli- J. Garnica et al. [19] proposed an architecture for URL in 100 GbE networks. Their architecture consists of TABLE II THE MACHINE INSTRUCTIONS OF FARIS opcode operand operation -
'"3*4 '"TUVOJGPSN3FTPVSDF*EFOUJpFS4QFDJpDpMUFS JUIBTJOTUSVDUJPOT SFHJTUFST BOEGSBNFTUBDL 'JMUFS3VMFT '"3*4 CZUFDPEFT '"3*4
QTFVEPNBDIJOF 63- SFTVMU NBUDIPSOPU -
'"3*4 .BDIJOF*OTUSVDUJPOT PQDPEF PQFSBOE PQFSBUJPO DIBS D JG41JTQPJOUJOHUPD JODSFNFOU1$BOE41PUIFSXJTF
JG UIFGSBNFTUBDLJTFNQUZ UIFOBCPSUNBUIJOH FMTFQPQ1$ BOE41GSPNUIFGSBNFTUBDL TLJQ@UP D JODSFNFOU41VOUJMJUQPJOUTUPDJGDXBTOPUGPVOE BCPSU NBUDIJOHPUIFSXJTFJODSFNFOU1$BOEQVTI1$ BOE 41UPUIFGSBNFTUBDL TLJQ@TDIFNF JG41JTQPJOUJOHUP63-TDIFNF JODSFNFOU41VOUJMJUJTOPU QPJOUJOHUP63-TDIFNFPUIFSXJTF BCPSUNBUDIJOH NBUDI pOJTINBUDIJOHTVDDFTTGVMMZ -
'"3*4 $PNQJMBUJPO3VMFT JOQVU JOTUSVDUJPO D TLJQ@UPD ? TLJQ@UPseparator D
DIBSD ? DIBSseparator ccPGUIFCFHJOOJOHPGBMJOF DIBShead TLJQ@TDIFNF cPGUIFCFHJOOJOHPGBMJOF DIBShead cPGUIFFOEPGBMJOF DIBStail -
'"3*4 &YBNQMF ccXXX BE?QIQ DIBSIFBE TLJQ@TDIFNF DIBSX DIBSX DIBSX
TLJQ@UPB DIBSE DIBSTFQBSBUPS DIBSQ DIBSI DIBSQ NBUDI IUUQFYBNQMFDPN IUUQXXXFYBNQMFDPNBEQIQ BCPSUIFSF NBUDI -
&WBMVBUJPO %"5"4&5 DBQUVSFSFBMUSB⒏DPG8*%&DBNQTQSJOH SBOEPNMZDIPTF 63-TGSPNUIFIUUQUSB⒏D TABLE V
ANTS OF WIDE CAMP 2015 SPRING Japanese non-Japanese nt non-student non-student total 26 67 9 102 5 1 0 6 31 68 9 108 m jp m jp m m jp jp jp jp m jp m jp et rg Fig. 3. Histogram of URL Length (1,000 Samples) IJTUPHSBNPG63-MFOHUI TBNQMFT -
&WBMVBUJPO &OWJSPONFOU $16 $PSFJ*)2<()[> $PSF)5 UVSCPCPPTU<()[> 04 .BD04
$ DPNQJMFS MMWNDMBOH -
&WBMVBUJPO .FNPSZ6TBHF CPU Core i7 I7-4870HQ 2.5
[GHz], 4-core, HT turbo boost 3.7 [GHz] OS MacOS 10.10.2 C++ compiler llvm clang++ 6.0 all typical easylist memory usage [MB] 0 175 350 525 700 hashed FARIS FARIS RE2 Irregexp (V8) typic easyl only 12.5% JavaScript w 2) Throug of our throug engines. Wh DPOTVNFPOMZNFNPSZJODPNQBSJTPOXJUI UIFDPOWFOUJPOBMJNQMFNFOUBUJPO '"3*4WT7 pMUFSpMFT pMUFSpMFT pMUFSpMFT -
&WBMVBUJPO .BUDIJOH5ISPVHIQVU 1,511 352 24,471 84 1,102
2,856 1,645 72,424 652 2,548 4,320 3,645 119,289 1,066 4,068 1,657 1,221 47,436 438 1,596 77,139 44,725 1,529,108 6,671 59,131 T ] all typical easylist #URLs per second 1 100 10000 hashed FARIS FARIS RE2 Irregexp (V8) BCPVUUJNFTGBTUFSUIBO UIFDPOWFOUJPOBMJNQMFNFOUBUJPO '"3*4WT7 pMUFSpMFT pMUFSpMFT pMUFSpMFT