We’ve had a full day operating with the CloudFlare CDN for gnuradio.org.
This hasn’t been long enough for the new DNS IP addresses to propagate
everywhere but we’re seeing most of our non-spider traffic coming
through
the CDN now.
The statistics from CloudFlare show that, for non web-crawler traffic,
in
the last 24-hours:
59,300 page visits
1881 unique IP addresses
6 GB of bandwidth used
The web crawler traffic from Google, Baidu, and others adds about 50%
more
hits on top of that from another 149 unique IP addresses.
So far, from what we’ve gathered from our server monitoring, the CPU
usage
has gone from a consistent 80% - 100% down to 30% - 70%, and the
bandwidth
we’re being charged for has decreased over 40%. You can judge for
yourself
the positive impact this has had on the web server response time.
That part of the experiment has gone well.
We’ve had to change the URL for access to the gnuradio.org Git
repository
when accessing it over the native Git protocol, as gnuradio.org now
resolves to one of the CloudFlare CDN servers. We’ve added a DNS entry
to
create ‘git.gnuradio.org’, which will always point directly to the
server.
Thus, the URL is:
$ git clone git://git.gnuradio.org/gnuradio
However, most users are accessing the repository over HTTP (such as when
using Marcus L.'s build-gnuradio script) using:
$ git clone gnuradio.git - GNU Radio
Not only will this continue to work, but CloudFlare is caching the data,
so
our web server never sees most of this traffic, and the overall
repository
checkout time should be going down.
We’ll look at the statistics again after a week to see how things go.
If
all is well we may even move to more aggressive caching from CloudFlare
(we’re on the ‘medium’ setting.)