ctFS

ctFS is a file system that Replaces File Indexing with Hardware Memory Translation through Contiguous File Allocation for Persistent Memory.

Code author: Robin Li

Access to the paper and presentation

It's a prototype and meant to prove the concept and demonstrate. I will maintain it and fix bugs progressively. You are welcome to test it and leave me feedback. Any contribution is much appreciated. You are also free to change and reuse the code as you wish.

Requirement

• Hardware: Intel Xeon CPU equiped with Intel Optan NVDIMM
• Software: Ubuntu 20.04 with all Kernel compile essentials

Tested environment

• System: ThinkSystem SR570
• CPU: Intel(R) Xeon(R) Silver 4215 CPU @ 2.50GHz
• DRAM: 6 x 16GB DDR4 RDIMM
• PMEM: 2 x 128GB Intel Optane DCPMM
• Kernel: 5.7.0-rc7+
• Distribution: Ubuntu Server 20.04 LTS

How to use:

Kernel part source code

1. Clone the Linux kernel and checkout v5.7-rc7
2. Copy the files in /kernel/* to the kernel repo kernel/drivers/dax/
3. Add the following code in mm/memory.c:

   EXPORT_SYMBOL_GPL(__pte_alloc);
   EXPORT_SYMBOL_GPL(__pmd_alloc);
   EXPORT_SYMBOL_GPL(__pud_alloc);
   EXPORT_SYMBOL_GPL(__p4d_alloc);

   After the end of these functions:

   __pte_alloc
   __pmd_alloc
   __pud_alloc
   __p4d_alloc

Compile

1. Compile and install the kernel
2. Compile ctFS user part (ctU):

   mkdir bld
   make

3. Compile mkfs

   cd test
   make mkfs

4. Compile fstest

   make fstest
   cd ..

Configure

1. Reboot to the compiled kernel and configure the PMEM:

   script/set_nvdimm.sh

2. Format

   test/mkfs 1

   Note that if you only need to format the ctU (user structure), run mkfs without any arguments. Running mkfs with any arguments will reset the ctK (kernel structure).

Run

1. Run with TEST_PROGRAM:

   script/run_ctfs.sh TEST_PROGRAM

2. Test with fstest:

   cd test
   script/run_ctfs.sh test/fstest -a -p "\\/test"

   Run with -h to show the help of fstest. The path to ctFS must start with "\", otherwise it will be bypassed to the regular file system.

Contact

Please feel free to reach me: [email protected].