CloudFlare : Java Glossary

Introduction

A company that offers a number of services of interest to people who have websites. Some are free and some cost a month. For free, they will provide Always Online. If your site goes down, their server takes over and displays a snapshot of your home page, with an explanatory banner. Also for free, they will mirror the static content of your site and serve it to your customers from the nearest mirror. The catch is the static content is updated only every day. For updating every 15 minutes, you have to pay. It does dozens of other things, many of which I have no clue as to their purpose.

Distributed Caching

It also offers CDN (Content Delivery Network) services to cache and serve your static contents from servers all over the world.

CloudFlare offers four types of service plans: Free, Pro per month, Business and Enterprise . Credit cards only No PayPal.

It will cache your static content at servers around the world to half response time. No matter what the traffic, your bill stays the same. You continue to use your same ISP (Internet Service Provider). You do not install any CloudFlare software on your server. You do not upload content to CloudFlare. You continue to serve your dynamic content direct from your own server.

With the free level, they spider your entire site once a week to prepare a complete cache to use in case your site goes down. You get automatic gzipping even if your server does not support it.

With pro level, you get prefetching so that the first hit is as fast as subsequent ones, priority over free traffic and SPDY (Speedy) support. With business level you get railgun, a super compressor that takes advantage of the fact most dynamic content changes slowly over time, and even updated static content is mostly the same. Basically, it compresses deltas.

Advantages

• Users around the world will get snappy response time.
• Search engines spidering your site will get snappy response and their spidering will not put a heavy load on your server.
• It will take the mother of all DOS (Denial of Service attack) attacks to take you down.
• Very little disruption.
• You don’t have to install any software at all on your server. However, if you sign up for the business level and use railgun, you need to install its compression engine on your server to rapidly get content from your site to CloudFlare.
• If your site goes down, popular pages will continue to work and unpopular ones will get a site is down page.
• The support people respond quickly and thoroughly. So many companies today don’t even read their queries. They just dispense pages from the website the users have already read based on noticing a few keywords.
• It will implement IPv6 (Internet Protocol Version 6) even if your server does not, so people can visit your site with IPv6.
• FTP (File Transfer Protocol) is rather flat-footed about uploading files that your users are downloading. The upload simply fails. With caching, your server is less busy so FTP has a better chance of finding a file it is trying to upload idle.

Disadvantages

• No PayPalNo cheques. No Money orders. No prepayment. No cash (is that legal?). Credit cards only. This is a show-stopper for me. This reminds me of a SciFi story I read as a teenager. In it the hero goes into a store and wants to pay cash. The clerks are horrified. To them, cash implied criminal enterprise. One offers to take the cash at a steep discount and put the purchase on his personal credit card. The incident seemed extremely strange at the time.
• With the pro level you can get SSL (Secure Sockets Layer), but the end user sees CloudFlare’s SSL certificate, not yours.
• Your server does not see hits of pages served from cache. It also sees pre-fetches as if they were hits from real people. This means your server’s page-hit statistics are not meaningful. If you sign up for the enterprise level, you get raw transaction logs from which you can compute individual page hit statistics, with regional breakdowns even, but not otherwise. If you must have accurate page-hit counts you will need some scheme that embeds JavaScript in your pages and reports back to some server when the end user receives the page from cache. My experience with such services is they greatly slow down response time, negating the effect of CloudFlare. This is discouraging me from trying even the free service.

Traditional Caching

Even if you don’t use CloudFlare, some of your pages may be cached by anonymous, transparent, caching servers. Your server puts a cache timeout in the header of the pages it sends out. Let’s say it said This page is good for one hour. After that, you had better get a fresh copy. Basically two unpleasant things can happen:

1. The hour expires, the page is still the same as ever, but the cache does an unnecessary fetch of a fresh page (or at least does a probe).
2. After 15 minutes, the page is updated. But the cache keeps serving the old version for another 45 minutes.

To succeed, your server needs ESP (Extra Sensory Perception) to guess when each page will next be updated. The essential problem is the cache protocol provides no way for the server to notify cachers that a page has been freshly updated and they should invalidate their cached copy, or that a page who timeout has expired still has not changed.

CloudFlare Caching

CloudFlare can cache using the same tools as any other cacher, however, it has some extra tools available to control it.

1. At the CloudFlare website, Clear the cache for the URL after updating the file, this will take effect worldwide within seconds. You would need to do this once for each file, each time you update. This would not be practical for routine use, unless you automated the process. You would use this if you had accidentally uploaded some embarrassing page and wanted it replaced immediately.
2. Enable development mode. This will turn off caching for your entire site and invalidate caches world wide for all your files.
3. Have CloudFlare cache such files for a shorter time and let them expire naturally. You can do this using Page Rules for the URL (Uniform Resource Locator) pattern you want to cache. This overrides the timeouts your server is applying, at least for CloudFlare. Note you can also use Page Rules to tell CloudFlare not to cache at all.
4. Have your server return shorter Cache-Control max-age values or Expires headers in the HTTP (Hypertext Transfer Protocol) Response. This would apply to all caching not just that done by CloudFlare.

So, in summary, cache expiry is controlled three ways:

1. By traditional cache timeout headers applied by your server.
2. By overriding the cache timeout rules at CloudFlare.
3. By forcing a global cache invalidate for a particular file at the CloudFlare website.

What CloudFlare could do in future is allow you to send them a list of files you, or give you a utility you can feed them to. It would then broadcast the invalidate requests, filtering out the invalidates that would happen naturally within 10 minutes.

One slightly awkward thing, you must persuade your ISP to reconfigure the DNS (Domain Name Service) name servers to use CloudFlare’s.

Requests for Change

• Accept forms of payment other than credit cards.
• Provide some way to get accurate per-page hit counts at the pro and business levels.
• Use long cache timeouts with a way to submit files just uploaded to the server to invalidate. I have some static content that goes for months at a time unchanged.

Summary

The documentation and support are exemplary. It is not only clear, it is aethesthetically laid out. The people who designed this website are true pros. They are also willing to explain how it works. I am very impressed by these people. There are so many companies now who consider their customers as unavoidable nuisances.