Performance bottleneck of pbft

Content

Why do I need 2f+1 messages (including myself and others) to judge the prepare certificate and commit certificate
Since we assume N=3f+1, Q=2f+1. See above about Quorum

Performance bottlenecks (all mentioned in the original paper)
View Change
Performance pressure is in Primary and the load is uneven
The cryptographic signature algorithm can be replaced by MAC (the MAC under normal circumstances cannot be used as a proof during View change, what should I do?)
Read operations are handled separately
Pre-execution response
Refer to learn this: http://ug93tad.github.io/pbft/

In-depth information
Original paper
http://ug93tad.github.io/pbft/ or http://cncc.bingj.com/cache.aspx?q=pbft+commit+phase&d=5047743736251098&mkt=en-US&setlang=en-US&w=fmNNrN7x4zj3cpsIdM7SLbG1Op2CFrEb
frabic-0.6 original code
http://www.pmg.csail.mit.edu/papers/bft-tocs.pdf
https://www.comp.nus.edu.sg/~rahul/allfiles/cs6234-16-pbft.pdf
https://www.cs.utexas.edu/~lorenzo/corsi/cs380d/past/15S2/notes/week14.pdf
http://www.cs.utah.edu/~stutsman/cs6963/public/pbft.pdf