Only leader can handle the change request, so does the leader would be the bottleneck?

[yucheng jing]

Recently, I research the raft algorithm.

The leader election and log replication confirm the consistency.

But if only the leader can receive the change request, so does the leader would be the bottleneck?

Is there a best solution?

Thx!

[Philip Haynes]

Dear Yucheng,

At a certain level, yes the leader is the bottleneck. But this is not the whole story.

In a replicated state machine either the state machine, or the result from the state machine, can be replicated with

the later being more common. In the former circumstance the amount of data processed by the consensus system

is not large (~256 bytes in our case) and is a useful system optimisation.

In parts of the high performance Java community, a common theme is to highly optimise a single control thread,

an approach described in the Disruptor concurrency component.  A strong leader makes it much more tractable to

to do this so that the consensus sub-system itself is not the net performance constraint for the system as a whole.

Kind Regards,

Philip

[Henrik Ingo]

Hi Jing

The answer is that Raft is not designed to solve the problem of scaling, it is designed to solve the problem of high availability. A Raft cluster will (roughly, and at best) have the same performance as a single server processing the same state machine would.

Raft can however be combined with techniques for scaling, such as sharding, and will be rouchly independent of such solutions.

henrik

--
You received this message because you are subscribed to the Google Groups "raft-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to raft-dev+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

henri...@avoinelama.fi
+358-40-5697354        skype: henrik.ingo            irc: hingo
www.openlife.cc

My LinkedIn profile: http://fi.linkedin.com/pub/henrik-ingo/3/232/8a7