Return to BSD News archive
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!simtel!news.sprintlink.net!howland.reston.ans.net!usc!nic-nac.CSU.net!csulb.edu!info.ucla.edu!nnrp.info.ucla.edu!laika!johnh From: johnh@ficus.cs.ucla.edu (John Heidemann) Newsgroups: comp.unix.bsd.bsdi.misc Subject: Re: Round Robin DNS?? Date: 17 Jul 95 20:04:57 GMT Organization: University of California, Los Angeles Lines: 52 Message-ID: <johnh.806011497@laika> References: <3uai48$2bq@lace.Colorado.EDU> <3uect4$mf8@park.uvsc.edu> NNTP-Posting-Host: laika.cs.ucla.edu X-Newsreader: NN version 6.5.0.b3.0 #11 (NOV) Terry Lambert <terry@cs.weber.edu> writes: >sryashur@sprint.uccs.edu (Surf-Kahuna) wrote: >] >] Ok, here's the concept: You have say, 3 machines all running >] a web server. (Let's assume your company is *SO* profitable >] that you actually need all three to handle the load) >] >] The problem: how to effectively ballance out the load of the >] WWW trffic among all three machines. >] >] The solution (at least one of them): a DNS server that >] alternates between the three IP addresses of the machines >] everytime it is queried for say, www.mycompany.com. >] ... > >This assumes (wrongly) that there will be a constant average >session length between servers. > >This would *probably* work adequately *if* you didn't end up >with a local cached copy of the host address that then gets >reused on your machine without "benefit" of the round robin. > >The *real* problem is that the protocols are designed to attach >to hosts instead of services. Consider: do you really give a >damn *where* the next HTML page comes from so long as it arrives? Terry is right, of course it would be better to do load balancing at a higher level. He also understates the problems: session lengths have odd distributions (probably bimodal) and many DNS servers cache names/address translations *after* the RR translation. However, for web sites with sufficiently large numbers of hits, the law of large numbers states that things should even out. RR DNS does HTTP load balancing pretty well in practice. Check out the following refernce for what they do at NCSA: @Unpublished{Kwan95a, author = "Thomas T. Kwan and Robert E. McGrath and Daniel A. Reed", title = "User Access Pattersn to {NCSA's} World Wide Web Server", note = "from web", year = 1995, keywords = "WWW, trace", url = "http://www-pablo.cs.uiuc.edu/Papers/WWW.ps.Z" } [This article may have been published by now. If so, let me know where.] -John Heidemann