OT: why are LAMP sites slow?
Skip Montanaro
skip at pobox.com
Thu Feb 3 23:12:14 EST 2005
Paul> I'm talking about the very familiar experience of clicking a link
Paul> and then waiting, waiting, waiting for the page to load. You
Paul> rarely see that happen with Ebay or Google. It happens all the
Paul> time with Wikipedia.
It's more than a bit unfair to compare Wikipedia with Ebay or Google. Even
though Wikipedia may be running on high-performance hardware, it's unlikely
that they have anything like the underlying network structure (replication,
connection speed, etc), total number of cpus or monetary resources to throw
at the problem that both Ebay and Google have. I suspect money trumps LAMP
every time.
Just as a quick comparison, I executed
host www.wikipedia.org
host www.google.com
on two different machines, my laptop here on Comcast's network in Chicago,
and at Mojam's co-lo server in Colorado Springs. I got the same results for
Wikipedia:
www.wikipedia.org has address 207.142.131.203
www.wikipedia.org has address 207.142.131.204
www.wikipedia.org has address 207.142.131.205
www.wikipedia.org has address 207.142.131.202
but different results for Google. Laptop/Chicago:
www.google.com is a nickname for www.google.akadns.net
www.google.akadns.net has address 64.233.161.104
www.google.akadns.net has address 64.233.161.99
www.google.akadns.net has address 64.233.161.147
Co-Lo server/Colorado Springs:
www.google.com is an alias for www.google.akadns.net.
www.google.akadns.net has address 64.233.187.99
www.google.akadns.net has address 64.233.187.104
We also know Google has thousands of CPUs (I heard 5,000 at one point and
that was a couple years ago). I doubt Wikipedia has more than a handful of
CPUs and they are probably all located in the same facility. Google's
front-end web servers are clearly distributed around the Internet. I
wouldn't be surprised if their back-end servers were widely distributed as
well.
Here's a link to an IEEE Micro article about Google's cluster architecture:
http://www.search3w.com/Siteresources/data/MediaArchive/files/Google%2015000%20servers%20secrest.pdf
It was published in 2003 and gives a figure of 15,000 commodity PCs.
Here's one quote from the beginning of the article:
To provide sufficient capacity to handle query traffic, our service
consists of multiple clusters distributed worldwide. Each cluster has
around a few thousand machines, and the geographically distributed setup
protects us against catastrophic data center failures (like those
arising from earthquakes and large-scale power failures).
Skip
More information about the Python-list
mailing list