jsburbidge (![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png)
jsburbidge) wrote2011-07-12 11:20 am
jsburbidge) wrote2011-07-12 11:20 am
Tyranny of the Installed Base
This article on Google+ is genuinely interesting in terms of what it reflects more generally about large, established organizations.
Another instance, from personal experience:
I used to work for Reuters (now, Thomson-Reuters, but this was before the merger), which makes a large chunk of its money out of real-time financial data (and some from making their news data available on similarly-structured realtime feeds). Their low-level protocols are directly descended from a time when they provided blocks of text pre-set to fit a fixed font terminal (much as IBM punchcards reflected the terminal display sizes), and their data propagation protocols are also very, um, legacy-centric. They have basically stuck a large number of kludges on top of this foundation to be able to deliver structured data. It works, but architecturally, it's a mess.
A little over a decade ago, Reuters was a party funding a semi-independent company called Tibco, which has a much more up-to-date set of data dissemination protocols, and there was a major project to convert all of the creaky legacy architecture to a clean, more efficient one based on the TibRV low-level protocols, plus (I would guess) additional higher-level simplifications. This had quite a high profile.
Then, one day, all the technical people were called in for a large conference call. Reuters was abandoning the migration entirely. Even though the risks associated with a dependency on a kludgy and inefficient base were well recognized, the tyranny of the installed base made a migration too great a risk for the company to manage. The two companies continue to cooperate and integrate their services, but the full-scale conversion over of the legacy systems never went forward: much of Reuters data flow is still old-style SSL (not Secure Socket Layer, but a TLA for "Source Sink Library").
Another instance, from personal experience:
I used to work for Reuters (now, Thomson-Reuters, but this was before the merger), which makes a large chunk of its money out of real-time financial data (and some from making their news data available on similarly-structured realtime feeds). Their low-level protocols are directly descended from a time when they provided blocks of text pre-set to fit a fixed font terminal (much as IBM punchcards reflected the terminal display sizes), and their data propagation protocols are also very, um, legacy-centric. They have basically stuck a large number of kludges on top of this foundation to be able to deliver structured data. It works, but architecturally, it's a mess.
A little over a decade ago, Reuters was a party funding a semi-independent company called Tibco, which has a much more up-to-date set of data dissemination protocols, and there was a major project to convert all of the creaky legacy architecture to a clean, more efficient one based on the TibRV low-level protocols, plus (I would guess) additional higher-level simplifications. This had quite a high profile.
Then, one day, all the technical people were called in for a large conference call. Reuters was abandoning the migration entirely. Even though the risks associated with a dependency on a kludgy and inefficient base were well recognized, the tyranny of the installed base made a migration too great a risk for the company to manage. The two companies continue to cooperate and integrate their services, but the full-scale conversion over of the legacy systems never went forward: much of Reuters data flow is still old-style SSL (not Secure Socket Layer, but a TLA for "Source Sink Library").