I don't have a detailed comparison, but I can tell you that the way I see it, Kafka implements both distributed logs and a pub-sub layer that uses these logs. BK is just the distributed logs and you need something like Hedwig (shipped here), DistributedLog (Twitter implementation), or ManagedLedgers (Yahoo! implementation) to work with it. You can also develop your own layer that is specific to your application.
BK also provides different durability and performance. BK syncs to disk before returning an add operation to a ledger (our logs) and uses separate devices to limit the interference between read traffic and write traffic. Write traffic consists of writing to ledgers and read traffic consists of reading from ledgers. I agree that a write-up comparing them would be useful to many people. -Flavio > On 12 Nov 2015, at 09:38, Yumei Li <[email protected]> wrote: > > Hello bookkeepers: > > I am a newbie to the bookkeeper community. We are looking for a replicated > log service that could be used for a pub/sub project. Looking around > bookkeeper documents, it sounds like bookkeeper is a pretty good fit for our > user cases: we need durability and could be able to scale up to ten thousands > of topics. But comparing to Kafka, it seems the API is pretty low level, we > have to manage the numbered ledgers. > > Do you guys have a detailed comparison between Kafka and BookKeeper? Like > architecture, features and performance. That would be a good guideline for > people like me coming to bookkeeper world. > > - YL > >
