Foreword
In2009,HTTP/1.1waswelloveradecadeold,andarguablystillthemost
popularapplicationprotocolontheinternet.Notonlywasitusedforbrowsing
theweb,itwasthego-toprotocolforamultitudeofotherthings.Itseaseofuse,
broadimplementation,andwidelysharedunderstandingbydevelopersand
operationengineersgaveithugeadvantages,andmadeithardtoreplace.Some
peoplewereevenstartingtosaythatitformeda“secondwaist”fortheclassic
hourglassmodeloftheinternet’sarchitecture.
However,HTTPwasshowingitsage.Thewebhadchangedtremendouslyinits
lifetime,anditsdemandsstrainedthevenerableprotocol.Nowloadingasingle
webpageofteninvolvedmakinghundredsofrequests,andtheircollective
overheadwasslowingdowntheweb.Asaresult,awholecottageindustryof
WebPerformanceOptimizationstartedformingtocreateworkarounds.
TheseproblemswereseenclearlyintheHTTPcommunity,butwedidn’thave
themandatetofixthem;previouseffortslikeHTTP-NGhadfailed,andwithout
strongsupportforaproposalfrombothwebbrowsersandservers,itfeltfoolish
tostartaspeculativeeffort.ThiswasreflectedintheHTTPworkinggroup’s
charteratthetime,whichsaid:
TheWorkingGroupmustnotintroduceanewversionofHTTPandshould
notaddnewfunctionalitytoHTTP.
Instead,ourmissionwastoclarifyHTTP’sspecification,and(atleastforme)to
rebuildastrongcommunityofHTTPimplementers.
Thatsaid,therewasstillinterestinmoreefficientexpressionsofHTTP’s
semantics,suchasRoyFielding’sWAKAproposal
1
(whichunfortunatelyhas
neverbeencompleted)andworkonHTTPoverSCTP
2
(primarilyatthe
UniversityofDelaware).
SometimeaftergivingatalkatGooglethattouchedonsomeofthesetopics,I
gotanotefromMikeBelshe,askingifwecouldmeet.OverdinneronCastro
StreetinMountainView,hesketchedoutthatGooglewasabouttoannouncean
HTTPreplacementprotocolcalledSPDY.
SPDYwasdifferentbecauseMikeworkedontheChromebrowser,andhewas
pairedwithRobertoPeon,whoworkedonGFE,Google’sfrontendwebserver.
Controllingbothendsoftheconnectionallowedthemtoiteratequickly,and