Alibaba Cloud Preemptible Instance Price Traces

The data

The Alibaba Cloud Preemptible Instance Price Traces is a data set of price histories spanning over almost 3 years, from November 2018 to July 2021. The files are named according to the instance type, the region and the avalability zone. There are 14,426 preemptible instance price history trace files. Each trace file contains price history data of precisely one preemptible instance. They are associated with the 19 regions that were available to us. Five regions were unavailable to us and thus were not analyzed: cn-wulanchabu, cn-heyuan, cn-guangzhou, cn-chengdu, and cn-nanjing. The data contains some gaps, most of them are due to the fact that Alibaba claims to export a month-long history whenever requested, but in practice some of the data we harvested was missing.

File Format

Each file is a 4-column csv file in the following format:

OriginPrice - The reference price of the same merchendise, as sold without bidding (on-demand), which was in effect at that time.
SpotPrice - The price which instance owners paid from this timestamp on.
TimeInSeconds - The timestamp in seconds. This is the only column we added, for convenience. It reflects the contents of the next column.
Timestamp - as provided by Alibaba Cloud.

For example:

OriginPrice,SpotPrice,TimeInSeconds,Timestamp
0.786,0.142,28860002,12/01/2018 0:40
0.786,0.145,28874702,12/01/2018 4:45
0.786,0.142,28889403,12/01/2018 8:50

Data Access and Citation

The traces are available in two ways. Both begin on November 13th, 2018.

The raw traces are available from alibaba.raw.tar.gz. This data includes all data until January 5th, 2022: over 3 years.
In addition, the cleaned traces, where a price is not repeated if it is announced when it is already the price, are available from alibaba.clean.tar.gz. This is the data analized in the paper, and it lasts until July 2021.

When using the traces, please refer to this page as the source of the data, so that others who read your work can find the traces and read the explanation. In addition, please read and cite the paper which describes the traces and exposes their features:

@inproceedings{preemptible,
      author = {Davidow Movsowitz, Danielle and Agmon Ben-Yehuda, Orna  and Dunkelman, Orr},
      title = {Deconstructing Alibaba Cloud's Preemptible Instance Pricing},
      year = {2023},
      publisher = {Association for Computing Machinery},
      address = {New York, NY, USA},
      booktitle = {Proceedings of The 32nd ACM International Symposium on High-Performance Parallel and Distributed Computing (ACM HPDC'23)},
}

We further ask that if you publish using these traces, you inform us so that we can link to your paper. Please contact Danielle Movsowitz Davidow at dani.movso (on gmail), Orna Agmon Ben-Yehuda or Orr Dunkelman regarding this page.

Publications using the Alibaba Traces

"Deconstructing Alibaba Cloud's Preemptible Instance Pricing", Davidow Movsowitz, Danielle and Agmon Ben-Yehuda, Orna and Dunkelman, Orr. HPDC'23.
"Sharp behavioral changes in preemptible instance pricing", Davidow Movsowitz, Danielle and Agmon Ben-Yehuda, Orna and Dunkelman, Orr. Extended Abstract. SYSTOR'22.