Question to the design of disk manager

[JigaoLuo]

I would like to ask a simple question: Is disk manager in bustub thread safe?

I would try to give my answer: it is not thread safe.

• reason 1: it has some int attributes.
• reason 2: std::fstream is not thread safe, refering to C++ Standard

27.1.3 Thread safety [iostreams.thread-safety]

Concurrent access to a stream object [string.streams, file.streams], stream buffer object [stream.buffers], or C Library stream [c.files] by multiple threads may result in a data race [intro.multithread] unless otherwise specified [iostream.objects]. [Note: Data races result in undefined behavior [intro.multithread].

---

Following words are under the assumption, that disk manager is not thread safe. If the assumption is not the truth, please ignore the following words.

Then I would like to ask: why to design the disk manager not to be thread safe?

Yes. We can have a mutex outside of the disk manager doing the protection. In the Project 1, we should have a mutex to protect the disk manager inside of buffer manager.

There is a way maybe better than that: to use POXIS pread, pwrite to implement the ReadPage and WritePage functions. Since the page_id represents a unique part of a file and different page_ids share no overlap in file. POSIX pread, pwrite would be thread safe and easy to
be implemented. (Possible disadvantage: not modern C++.)

I am hoping and being grateful to see an answer. 😄

---

The motivation of the question:
I want to have low lock(mutex) granularity in my implementation: the buffer manager have to release it's global_mutex during ReadPage and WritePage these I/O operations. Then I came to this question and did some reading. I even have upload a version to gradescope and passed the concurrency test, in which the disk manager is not protected by any mutex.

[lmwnshn]

Just general comments:

I didn't write the original disk manager, so I'm not sure what the original design motivations were.

There was a version of the buffer pool project where we had a scoreboard for buffer pool performance. You might want to try writing and running some local benchmarks for fun.