Hello Reza and Stefano, Thanks for the emails. Where I work we have implemented a device detection pattern based on prioritizing data sources. There's certain properties that are interesting to us, so we build "dictionaries" that map those properties to the data source. These data sources so far are UAProf and WURFL. I'll give OpenDDR a try and prioritize it over WURFL.
UAProfile has yielded favorable results over WURFL, we have seen a 30% increase in the accuracy of our detection, and a significant increase in performance doing back-end and front-end device identification. I would like to know, if possible, what typical devices you receive traffic from. We're currently operating in South Africa, where we see plenty of legacy and low-end devices, which makes proper identification tougher. One of our most popular is a Nokia C5. It's also important to point out that UAProfile is very good, although it has its drawbacks, one of them being that some of the links provided are unavailable. Also, the XML files provided can contain errors, and there's no syntax agreement on how to denote certain properties. For example, some companies specify the screen size as "320x240" while others specify it as "320*240", so your parsers need to be tolerant. Thanks for the emails again. Regards, On Thu, Aug 16, 2012 at 8:47 PM, Naghibi, Reza <[email protected]> wrote: > Im very happy with the OpenDDR dataset. I would easily put the accuracy in > the high 90% area and its only getting better with time. The accuracy is > due to the fact that OpenDDR identifies devices using pattern matching. > This is a highly effective and simple technique and gives it a huge > advantage over other DDR products. For example, I evaluated WURFL and found > it had subpar accuracy, poor performance, and a bloated DDR. > > The monthly releases are very "fresh". Been using OpenDDR since late 2011 > and haven't had any devices slip thru. I would make sure you update your > DDR several times a year to stay ahead of any new devices. Device discovery > will obviously improve as these projects get more visibility. > > Reza > > -----Original Message----- > From: Stefano Andreani [mailto:[email protected]] > Sent: Thursday, August 16, 2012 12:25 PM > To: [email protected] > Subject: Re: device routing and feature detection > > Hi Reza, > > Thanks for sharing this: there is still a lot of confusion in the market > about device detection, and your classification is very clear and > meaningful. > > TheWeatherChannel has a large set of devices accessing its contents, so > you should have a good visibility about accuracy of the dataset under use. > What is your opinion on the accuracy of OpenDDR dataset? Are the monthly > updates "fresh" enough to allow a correct mapping of new devices accessing > TWC web sites? > > Thanks, > Stefano. > > On 13/ago/2012, at 22.32, Naghibi, Reza wrote: > > > I recently had an interesting conversation around device detection and > decided to share it with the list. I categorize device detection into the > following: > > > > -Backend device detection and request routing > > Looking at the HTTP User-Agent field allows requests to be > accurately routed to device specific sites (or services or resources). This > gives you a lot of design flexibility. You can go ahead and target a given > site to devices where it's most appropriate and not have to worry about how > to make that design work across a wider array of devices. This has many > performance benefits, both backend, network, and frontend. > > > > -Client side feature detection > > This represents the other side of the device detection spectrum. > What is important here is that you detect the presence (or lack of) > features and not the devices themselves. This is done very differently than > backend device detection and is usually accomplished with CSS and > Javascript. Feature detection on the client side will provide for a much > more responsive design across devices than device detection. > > > > -Client side device detection > > I found client side device detection only helpful in limited > scenarios like metrics, ad targeting, and client side analytics. Client > side device detection can be done by exposing backend device detection as a > JSON service and having the browser make a call to the service. > > > > I found the most value in combining backend device detection routing > with client side feature detection. This allows sites and services to be > designed with specific device classes in mind (feature phones, smart > phones, tablets, desktop browsers) which allows for the reduction of > responsiveness complexity and a design which is more native to the target > device. The key here is to be flexible. Purely responsive designs and > purely device detected designs are not flexible enough to meet all > performance and design demands. > > > > I recently open sourced our backend device detection framework which > uses OpenDDR and provides backend integration: > https://github.com/TheWeatherChannel/dClass. dClass is used for backend > device routing and for client side device detection service (both in > Varnish). > > > > Reza Naghibi > > > > > > > > -- Carlos D'Agostino.
