hustaisgroup / cdbtune Goto Github PK

[用两种语言写了提问,每个内容都是一样的。 为了避免传达错误,用2种语言填写了。]
[I wrote the question in two languages, and each content was the same. It was written in two languages to avoid miscommunication.]

回放 YCSB 结果发现， readproportion= 1 会提供 99th latency 作为结果值。
但是,像readproportion=0.95、updateproportion=0.05的情况一样,同时存在多个operation的workload分别输出read的latency值和update的latency值。

(e.g.)
[READ], 99thPercentileLatency(us), '@@@'
[UPDATE], 99thPercentileLatency(us), '###'

不计算全部workload的99th latency。

大家是如何确认整个workload的99th latency的呢？
确认代码后搜索，但是找不到，所以留下issue。

下面的英语文章为了应对上述问题内容传达错误的情况,用英语写了相同的内容。

As a result of turning the YCSB, it was confirmed that 99th latency is given as the result value in case of readportion=1.
However, as in the case of readportion=0.95 and updateportion=0.05, a workload with multiple operations simultaneously outputs the latency value of the read and the latency value of the update as a result of 99th latency, respectively.

(e.g.)
[READ], 99thPercentileLatency(us), '@@@'
[UPDATE], 99thPercentileLatency(us), '###'

The 99th latency for the entire workload is not calculated.

How did you check the 99th latency for the entire workload?
I'm checking the code of ycsb and searching it, but I can't find it, so I leave an issue

在ddpg.py文件中第287行，OUProcess代表的是什么？我没有找到相关实现，希望能够帮忙解答一下，谢谢。

I have a question about the paper, why the performance of CDBTune is much worse than other algorithm in figure 5?

NameError: global name 'OUProcess' is not defined???

Hello guys,

Thanks for the instruction and codes, I download the codes and run on my computer using local MySQL server, and I found there are some wrong with the MySQLEnv .

The initialize function has not been completed. So I am looking forward to see the complete implements.

This would be a great help for my scientific work.

Best regards,
Scott

Hi, thanks for sharing the code.
In /models/ddpg.py, Line 287,
self.noise = OUProcess(n_actions)

The definition of OUProcess is not found.

Can this project only run on Ubuntu version 16.04?

Hello together,

thanks for providing the paper; made for a very interesting read.
Is there any update on the open-sourcing-process of the CDBTune project?
This would be a great help for my scientific work.

Best regards
Michael

说明里的AutoTune和项目名CDBTune是一个吗？需要将AutoTune改为CDBTune吗？

ｓｏｒｒｙ　Ｉ　ｄｏｎ＇ｔ　ｓｅｅ　ｈｏｗ　ｏｔｈｅｒ＿ｋｎｏｂｓ　ａｄｄ　ｉｎｔｏ　ｋｎｏｂ＿ｄｅｔａｉｌｓ

Hi everyone,

I have some questions after reading this paper and CDBTune's code.

• Firstly, as the paper mentions "CDBTune collects 1500 samples", does it means DDPG uses about 2k training data to converge?
• Next, which process in CDBTune is paralleled by 30 servers?
• Finally, how possible does the tuned result achieve the "adaptability"? What is the tuned knobs parameter in CDB?

Hope you can answer me and push the official open source code as soon as possible.

Thank you.

README文档第18步，需要自行设置environment模块下knobs.py文件中各个knob的默认值。

如何设置？是每一个都添加上去还是随机选几个knobs啊？初始值又如何选定？

i am trying this project at my end but model is not getting converged can i get the trained weight for this model