The trace datasets are unavailable

[PureWhiteWu]

Dropbox shows that all three files are not available.
Could you please upload those files to the repo(maybe using git-lfs?)

https://www.dropbox.com/sh/9ii9sc7spcgzrth/z9qaItlfnw/Papers/ARCTraces/DS1.lis

[PureWhiteWu]

Or is there any other way I can get those datasets?
Thanks!

[tatsuya6502]

Hi. For now, I put all the ARC datasets I have to an Amazon S3 bucket. Please download them from here: https://arc-ds-2023.s3.ap-northeast-1.amazonaws.com/arc-dataset.txz (49.5 MB) (Deleted. See this comment for the new location)

Then expand the txz:

## NOTE: You might need to install `xz` (xz-utils) in addition to 
## `tar` to expand a `txz` file.

$ tar xf arc-dataset.txz

Later, I will put them to a (new?) Git repository with git-lfs.

I am redistributing them with a permission from Dr. Dharmendra S. Modha, a coauthor of the ARC papers. I contacted Dr. Modha after seeing your message last night, and he kindly allowed me to publish them with a reference to ARC papers.

A referent to an ARC paper:

Nimrod Megiddo and Dharmendra S. Modha, "ARC: A Self-Tuning, Low Overhead Replacement Cache," USENIX Conference on File and Storage Technologies (FAST 03), San Francisco, CA, pp. 115-130, March 31-April 2, 2003.

I do not have some of the datasets such as ConCat.lis. If I remember correctly, their download links were already broken when I tried to download few year ago. Anyway, the above arc-dataset.txz should have enough stuff to get started.

Also note that current version of moka has a large hard-coded buffer (diagram). So you should avoid to test moka with smaller datasets such as OLTP.lis (11 MB), otherwise you will get a skewed result with very high cache hit rate as the entire dataset may fit into the buffer.

[PureWhiteWu]

Thank you very much for this!

[tatsuya6502]

FYI, I moved the trace files here: https://github.com/moka-rs/cache-trace/tree/main/arc