[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Possibly moving Debian services to a CDN

This one time, at band camp, Ian Jackson said:
> I don't see a clear explanation of what the motivation is to switch to
> a commercial CDN.  Can you clarify ?  That will help us understand
> what we would be giving up if we decline to make this change.

There are probably several having to do with effort and effectiveness,
but I'll just share one story that's been ongoing.  At debconf in Spain,
I spoke with several people about the idea of putting in a security
mirror in the vicinity of China.  We are still talking about this, and
users in China have made do with .. I don't know what in the meantime.

Getting things closer to our users makes things nicer for them - time to
first byte can be appalling when viewing our web page from some parts of
the world, especially when on a mobile device.  Time to first byte
matters less when downloading large packages, but again, if we can
get caching with large bandwidth, it's a better experience for the end

I count at least a dozen d.o machines that are nothing but mirrors
of one thing or another, and we still don't have very good coverage.
There are lots and lots of subdomains (lintian, popcon, people, etc)
that are in basically one place.  That's a single point of failure,
and also, half the time, on the far side of the world from our users.

Sure, we could add more mirror machines, and we could make more and more
mirror infrastructure and build our own CDN.  At some point, though,
it makes sense to me to say, "other people already do this really well".
We can carry on the way we are, but I think we'll lose out on improving
throughput for our end users, getting close to users in parts of the
world that have been difficult for us to find hardware or hosting,
and reduce the workload on DSA while making more services more available.

I understand your concerns about privacy, although I think they are
perhaps unrealistic.  You're asking, in a world where security services
tap uplink cables and passively record metadata about all traffic, for me
to get upset about whether someone logs an IP or not.  Once the transit
is proven to be unsafe, the endpoints no longer matter, in my mind.

That being said, I don't think that at least the first iteration of
this should force anyone off of their preferred mirror.  I don't think
we want to rush into this, and I don't think we want to throw out all
the work we've done already to make our home-brewed CDN.  I think we
can let them coexist for a while, and see how we do.  We have had one
report that someone gets better throughput off of their existing mirror
than they do from one of the CDNs offering their support.  If we get
better coverage for a small number of users and worse coverage for most
of our existing users, that's clearly not a tradeoff worth making.

I guess what I'm saying is, I can see lots of ways that this can make
things better.  Right now, we don't have any metrics about the
tradeoffs, all we have are emotional responses.  I'd like to start
trying, so we can collect those metrics and make an informed decision.
We can always say, "thanks but no thanks" later, right?

|   ,''`.                                            Stephen Gran |
|  : :' :                                        sgran@debian.org |
|  `. `'                        Debian user, admin, and developer |
|    `-                                     http://www.debian.org |

Attachment: signature.asc
Description: Digital signature

Reply to: