alexandreio / deepcache_netai2018

DeepCache: A Deep Learning Based Framework For Content Caching

By: Arvind Narayanan, Saurabh Verma, Eman Ramadan, Pariya Babaie, and Zhi-Li Zhang

In this paper, we present DEEPCACHE a novel Framework for content caching, which can significantly boost cache performance. Our Framework is based on powerful deep recurrent neural network models. It comprises of two main components: i) Object Characteristics Predictor, which builds upon deep LSTM Encoder-Decoder model to predict the future characteristics of an object (such as object popularity) -- to the best of our knowledge, we are the first to propose LSTM Encoder-Decoder model for content caching; ii) a caching policy component, which accounts for predicted information of objects to make smart caching decisions. In our thorough experiments, we show that applying DEEPCACHE Framework to existing cache policies, such as LRU and k-LRU, significantly boosts the number of cache hits.

DeepCache is an open source version of the DeepCache platform described in our NetAI 2018 paper, released under the BSD license.

The available code now ONLY generates the synthetic and MediSyn datasets used in our paper.

Requests Generation:

• generateSyntheticDataset.py: This code generates a synthetic dataset which has multiple sessions where the object popularity and the request interarrival distribution changes from one session to another.

• generateMediSynDataset.py: This code is based on Tang, Wenting, et al. "MediSyn: A synthetic streaming media service workload generator." Proceedings of the 13th international workshop on Network and operating systems support for digital audio and video. ACM, 2003.

  The variables are mostly based on the paper, but we did some changes to generate requests which fit our need as shown in the request analysis plots. In addition to object popularity, and request interarrival distribution, this dataset simulates real traffic by:

  1. having new objects introduced at different times
  2. objects have different types with variable lifespans
  3. requests for each object are generated based on an hourly request ratio

Requests Analysis:

• requestAnalysis.py: analyzes the generated requests to generate the number of active objects, object frequency, object lifespan, object introduction day.
• plotObjectProperties.py: plots the previous data about object properties for the generated dataset.

Sample Datasets with Analysis Plots:

Synthetic Dataset Sample

MediSyn Dataset Sample

Citing DeepCache:

If you use DeepCache in your work, please cite our paper:

@inproceedings{Narayanan:2018:DDL:3229543.3229555,
	 author = {Narayanan, Arvind and Verma, Saurabh and Ramadan, Eman and Babaie, Pariya and Zhang, Zhi-Li},
	 title = {DeepCache: A Deep Learning Based Framework For Content Caching},
	 booktitle = {Proceedings of the 2018 Workshop on Network Meets AI \& ML},
	 series = {NetAI'18},
	 year = {2018},
	 isbn = {978-1-4503-5911-5},
	 location = {Budapest, Hungary},
	 pages = {48--53},
	 numpages = {6},
	 url = {http://doi.acm.org/10.1145/3229543.3229555},
	 doi = {10.1145/3229543.3229555},
	 acmid = {3229555},
	 publisher = {ACM},
}