thawk105 / ccbench Goto Github PK

Problem: Lack of acquiring lock around RLL at read phase.

Place:

Issue status: I will fix when I have enough time.

• after sigmod review, use likely, unlikely at include/compiler.hh and improve performance uniformly.

Google C++ スタイルガイドに従う．
https://ttsuki.github.io/styleguide/cppguide.ja.html

It‘s a excellent work for benchmark of concurrent control. I want to listen your presentation. Do you have youtube channel or somewhere else?

An Analysis of Concurrency Control Protocols for In-Memory Databases with CCBench is really a great paper.
After reading it, I have three questions which confuse me a lot. If you could give me some answers, I'll be very appreciated.

1. In your experiments, you count the number of committed and aborted transactions as the average throughput. Then you how can you determine that after using ReadPhaseExtension, the throughput improvement of CC is not caused by the large count of aborted transactions? And whether it's reasonable to include the number of aborted transactions in the throughput?
2. How do you measure the throughput of fetch_add? And how do you measure the latency breakdown of the Cicada operations?
3. How do you store the versions in your version chain? Are the versions organized in the new version to the old version way or in the old version to the new version way?

Mltiversion Concurrency Control (MVCC) is a family of widely-adopted CC protocols including MV2PL and MVOCC. Would be nice if those protocols are supported.

How did you evaluate the L3 cache miss in your experiments? If you could explain in detail, I'll be extremely appreciative of your kindness!!!

一部プロトコルは
procedure_list pro[---]
std::vector pro

のようなばらつきがあるので統一する．

トランザクション実行の for ループも同様（vector - iteration を使う．)

Could I ask a question that is all CC protocols implemented in CCBench run under the same isolation level? If they run under the same isolation level, could you tell me what it is. If they run under different isolation levels, could you tell me what they are respectively. I'll be very appreciated if you could answer my question! Thank you very much!!