Progress on Blink

[Daniel Wisehart]

Hello Gents,

How is your progress on Blink?  Have you started using it in different situations?  Now that you have used it for a while, what does Blink do very well?

Thanks,

Daniel

[Anders Furuhed]

Hi Daniel!

We are doing a lot with Blink internally and with clients. 

The work that we know others do independently has not been made public. We know of people that have decided to first use Blink internally before making public interfaces.

We have not received feedback indicating that Blink should be taken in a new / different direction, so our views leading to the design are just being reaffirmed. 

We have however had questions if Blink can be used to carry a mixed bag of key value pairs, and we have responded that the design is optimized for use where some care is spent in formulating messages. Such message design is the key to getting performance in the first place, so while Blink can be used to carry also a bag of tags, it is not optimized for that type of use.

There have been questions regarding the Native format, basically saying that we could make it go even faster by making more compromises in producer and consumer access patterns. We are planning to produce more benchmarks on the blog. So far, it seems to us that the Native format is a good fit for purpose relative to our claimed use cases.

We get a lot of requests in publishing an open source C++ implementation. It has been taking us some time to get there, as we want to make sure that we eat all dog food that we publish. We had some criticism that the Java implementation was not as highly optimized as would have been expected from something coming from us, and the obvious reason to that is that we have ourselves been using it for integrating purposes, on the edge instead of in our core. We have not taking the route of publishing a C++ version early. We have however made sure to get independent feedback on our C++ code to make sure that it works well outside our environment.

To summarize, we feel that Blink is stable specification wise. There could be a flurry of changes if others pick it up to be used in other settings. For instance, the type system is selected out of pragmatism, based on what we found to be sensible to us and others may want more or different types.

I have personally been doing work that is tangential to Blink in the form of a new session layer that we use together with Blink. That work has had no impact on Blink itself, but Blink has been a hugely helpful presentation layer in that stack, doing most of the heavy lifting and supporting all the types of encodings that I could wish for, without compromises.

We have had very little feedback or discussion through these public forums. More so in private, but we would welcome more discussion in public. I think the expectation is that we provide more open source implementation as the next step.

Kind Regards,

Anders