18 Nov
2012
18 Nov
'12
1:46 p.m.
On 17 November 2012 20:14, Daniel Snyder <ddsnyder at zoominternet.net> wrote:
Found this bit of info years ago, Altavista hardware.
----------------------------------------------------------
Newsgroups: alt.fan.cecil-adams
From: ehr... at his.com (Edward Rice)
Date: 1997/02/10
Subject: Re: Chinese restaurants; Alta Vista
The Hardware Behind AltaVista
AltaVista: AlphaStation 500, 256 MB memory, 6GB disk.
AlphaStation 500's handle all external traffic to the site.
They run a custom multi-threaded Web server which sends
queries to the Web indexer and News indexer.
Web Indexer: AlphaServer 8400 5/300, 10 processors, 6 GB
memory, 210 GB RAID disk. This model is the most powerful
computer built by Digital. These servers run the query
engine. The Web index is larger than 40 GB, but most
requests take less than a second.
Scooter: AlphaServer 4100 5/300, 1.5 GB memory, 30 GB RAID
disk. The super-spider runs from this machine. It fetches
pages from the Web and sends them to Vista, our primary web
indexer.
Vista: AlphaServer 4100 5/300, 2 processors, 2GB memory,
180GB RAID disk. This machine indexes Scooter output and
serves as a central distribution point for new index data.
News Indexer: AlphaServer 600 5/333, 896MB memory, 13 GB
disk. This machine keeps an up-to-date index of the news
spool: since new articles appear and old articles expire all
the time, it is in fact quite busy, even though the index it
serves is much smaller than the Web index.
News Server: AlphaServer 600 5/333, 896MB memory, 24 GB RAID
disks. It maintains a current news spool for the News
Indexer. It also serves the articles via http to those of
you who don't want to know about news servers but want to
read news.
Those aren't typos, either -- those are machines with many, MANY megabytes
of RAM, and they're hooked up to some of the fastest random-access backing
store in the world with capacities in the many, many gigabyte range.
AltaVista is fast because DEC threw some phenomenally capable resources at
the problem.
---------------------------------------------------------------------
Dan Snyder
Butler, PA
Remarkable specs for 1997. Really extraordinarily high. This must have
cost very serious money. No way Altavista ever paid for itself, not by
several orders of magnitude.
This is why the discovery of an efficient parallelising algorithm -
viz, Map/Reduce - was such an extremely big deal for Google.
Map/Reduce and PageRank made Brin & Page their tens of billions, and
for once, I don't begrudge them it at all. They earned it fair and
square through extreme brilliance, I think, unlike Gates or Jobs...
--
Liam Proven ? Profile: http://lproven.livejournal.com/profile
Email: lproven at cix.co.uk ? GMail/G+/Twitter/Flickr/Facebook: lproven
MSN: lproven at hotmail.com ? Skype/AIM/Yahoo/LinkedIn: liamproven
Tel: +44 20-8685-0498 ? Cell: +44 7939-087884