Hello,

Now that Linode has 6 datacenters, would a service for pull content distribution make sense?

This is a way of making the delivery of static content more closer to users.

Just a suggestion

Thanks

It might, but it's worth pointing out that there's nothing stopping you from doing this yourself. I don't mean to come across as negative, only to point out how powerful a platform Linode is, that you could easily design a CDN on top of it.

How exactly you're suggesting to use Linode as CDN?
I think it's better if Linode itself provide CDN service out-of-box. It would be great and efficient as all static content as well as dynamic content serving from one provider.
And that will give Linode another stream of revenue too.

_________________
Newbie to Linode....

Get a linode in each datacenter, synchronize the static content between each of them, boom, you have a CDN. You can direct users to a given content server by using a free geolocation database on their IP when they first hit the server, which should be good enough considering how far apart most of the datacenters are.

If you want to get fancier, a caching reverse proxy on the CDNs would automatically handle keeping only the most highly demanded content locally cached on each node, allowing the CDN machines to have a much smaller storage space than the master node. Even Squid can do this fairly trivially, and you don't have to worry about synchronization.

I'm not saying it wouldn't be nice if Linode did this all for us, I'm just saying that if there is a strong need, you can do it yourself relatively cheaply. The minimum monthly cost to run a Linode-based CDN with coverage from each datacenter is $101.76 per month, but you probably only need three or four Linode datacenters to get decent geographic coverage. Some US datacenters are close enough together that they're less valuable. It's more useful, for example, to have Newark/London/Tokyo than it is to have Newark/Atlanta/Dallas.

But that's lot of money and effort for relatively small blog having 75k visits monthly...
If linode can provide something out of the box, it would be much cheaper, efficient and painless.
Right now I'm considering MaxCDN, but looking at some of the reviews, I want to wait for Linode.
It would be great if some staff personal from Linode can comment on this.

-Divyang

_________________
Newbie to Linode....

Which begets the question what's the CDN for, then? CDNs are used when large amount of traffic justifies geographic dispersion of data. That large amount of traffic should then generate revenue to pay the infrastructure.

BTW, I don't like promoting Linode competitors on Linode forum, but there's one with "files (in the) cloud" service that I used for over a year, primarily as offsite backup. With a flip of a switch you can turn the bucket into public CDN, and it has nice Python bindings so I incorporated it into my apps directly, deliverying all uploaded content to it.

The drawback was (still is if I'm not mistaken) that you can't CNAME the buckets so if you want to switch the provider it'll be a huge pain to rewrite all the links to CDN'ed content.

So if Linode was to implement one, I'd say these would be awesome features:

1. RESTful api, with Python bindings ( )
2. Ability to select locations
3. Private/public bucket switch
4. CNAME'd or otherwise branded bucket URLs
5. Automatically scallable, ie. or at least storage upgrades with a flip of a switch

If you're changing the base URL in more than one place, you're working too hard. See also this sexy beast.

_________________

/* TODO: need to add signature to posts */

As I'm developing with Pyramid, it's a matter of single line change in config.

But I'm talking about stuff for example used in Facebook apps which was my case, or any other scenario where you have to distribute content to systems where you can't easily (or at all) change it. Fortunately the app died (lol!) before I needed to switch CDN provider, but still....

Unpopular websites aren't allowed to be fast?

By the way, if you're referring to the competitor I think you are, they offer CNAMEs for http content now, though not https.

_________________
Matt Nordhoff (aka Peng on IRC)

So it seems they do. Funny, I never saw a news post or any notification...

Rather, they don't need a CDN to be fast. Heck, I'm about as far as you can possibly get from Australia, and yet a site like ausgamers loads plenty fast for me. Yes, geographic distribution of static content will speed up loading a bit, but if you've got a well optimized and well connected site to begin with, not enough to be worth it.

CDNs tend to be more useful for the "handle massive load with minimal effort" use case, IMO.

@Guspaz

So smaller shops shouldn't be efficient and serve better experience?
I think otherwise. Maintaining quality of service has nothing do with your size of operation IMHO.
And yes I really think, CDN will improve speed of my blog.

-Divyang

_________________
Newbie to Linode....

CDN can help increase perceived pages speed yes. If it shaves down asset load time from say 300ms across the globe down to 30ms from geolocated nearby node.

Almost imperceptible for single request, but if your page was crawling with graphics and static whatnots, and browsers by default open 2 - 6 connections at once, that's 270msec for each group of 2-6 assets on the page.

Then the purpose of having a CDN is quite minimal. A NodeBalancer and an extra node in other common geographic areas would suffice quite well for this to distribute traffic. A simple rsync script could do the job.

Linode doesn't offer file system replication at this point, only block-level replication, so at this point, it would represent a totally new area; it's not as if this functionality is already offered. A block-level replication is not bandwidth, memory, or storage efficient for incremental content updates. Even if the volume were partitioned efficiently, an entire block-level replication to deploy a new version of a 3KB image file is not very sensible.