This dataset represents one week of cache requests over two east coast, two west coast, two asia, and two lower tier metals. Requests are sampled at a rate of 1/100 using uniform sampling.
The dataset contains the fields:
| name | data type | description |
|---|---|---|
time | uint32 | |
obj_id | uint64 | hash of url |
obj_size | uint32 | |
ttl | uint32 | |
age | uint32 | calculated as time-last-modified time |
extension | string | (enum) jpg, png, ts, html, js... |
origin_response_chunked | bool | the presence of 'Transfer-Encoding: chunked' in HTTP header |
hostname | uint64 | |
request_method | bool | only GET and PURGE requests |
colo | uint32 | used only in dataset 2 |
colo_tier | string | (enum) used only in dataset 2 |
Example csv file line:
1728492382, 38fe47fc1aa3c1973099107d87216057, 12098, 3600, 36, jpg, 1, hippos.in.love, 1, dtw01, 2
And an invocation example of the helper script:
more = True
while more:
if so_inclined: more = process_line()
be_done()
This dataset was used to generate the graphs for the paper
"FIFO queues are all you need for cache eviction", published in the Proceedings of the 29th Symposium on Operating Systems Principles.