NHA-Ext2 / News: Recent posts

We have tested our NHA-EXT2 on several SSD for a month.
- workstation: DELL PowerEdge T105
- workload: developer environment
- SSD: MTRON(MSP-SATA7035-064), Samsung(MCCOE64G)
There is no bug so far. We are monitoring NHA-EXT2 now.

At present, we are developing the NHA-cleaner, but it is difficult to develop the NHA-cleaner. So developing the NHA-cleaner may takes some time.

* plans
1. Researchs in the past survey
2. The architecture of NHA-cleaner design
3. Implementation of the NHA-cleaner on linux
4. Test and analysis

1. We faced some problem which was understanding the linux kernel page cache while we modified the source code of ext2 file system.

2. The present progressing rate of this project is 85% of total rate.

1. We have upgraded the allocation policy of the ext2 and experimented the performace of the system that applied the ext2.
2. The present progressing rate of this project is 70% of total rate.

1. We are studying the block layer of linux-kernel to implement segment writing.

2. We are making proto-type implementation of segments writing.

We are upgrading NHA-Ext2 allocation policy.
Current version of NHA-Ext2 shows slightly low performance.
Therefore modification should be required.
And we are also upgrading Postmark benchmark source for accurate experiments.

1. In order to implement New Hybrid Allocation policy,we have studied the delayed allocation and the page cache in the ext2 file system.

2. We faced some problem which was understanding the linux vfs layer while we modified the source code of ext2 file system.

* To optimize the SSD performance, filesystem have to write like the LFS(Log-Struct filesystem)'s allocation policy. So, we are modifying the allocation policy of ext2 filesystem as the LFS. We expect that the performance of modified ext2 will improve.

1. In order to implement NHA-Ext2, we have analyzed a both relationship the page cache in linux-kernel and the blocks of file-system.

2. We are applying the our analysis to design of NHA-Ext2.

Designing details of NHA-Ext2 allocation policy. We determined when file system utilization is low, NHA-Ext2 would allocate blocks like LFS but when utilization is high, NHA-Ext2 would perform like conventional Ext2 allocation policy.

1. Analyzing ext2, nilfs.
- ext2 : Pre-Allocation Method
- nilfs : Redirty technique

2. NHA-Ext2 needs to implement deleyed allocation technique

1. Analyzing the layout and allocation policy of ext2.
- We analyzed about the ext2 filesystem through studying the understanding the linux kernel and checking the source of linux kernel.

2. Studying LFS(Log-Structured File System) schme.
- We have completed it.

3. Experiment
- We made steady progress the experiments for checking the performance .
- HDD vs. SSD
- Ext2 vs. LFS

* The experimentation method of LFS used the NILFS(New Implementation Log File System) on Linux

1. Analyzing experiment result of SSD and HDD performance.
- In 1st week experimental, we have realized that sequential allocating of block is best way on SSD.
- So, we need to study sequential allocating scheme such as LFS.

2. Studying LFS(Log-Structured File System) scheme.
- Now studying it.

1. We compare SSD with HDD on linux machine and we got characteristics of both devices.
We observed that SSD performed well in any read request patterns but in case of write request, SSD showed high performance in sequential write pattern but low performance in random write performance.
But, HDD was not affected whether read or write request and pattern of I/O has been the most important factor of HDD performance. In case of sequential I/O, HDD showed high performance. In case of random I/O, HDD remained low performance.