Simulator of experiments presented in
Y. Li, T. Si Salem, G. Neglia, and S. Ioannidis, "Online Caching Networks with Adversarial Guarantees", ACM SIGMETRICS / IFIP PERFORMANCE 2022.
Please cite this paper if you intend to use this code for your research.
The following is an execution example:
python CacheNetwork.py res/fixed-popularity \
--graph_type dtelekom \
--cache_type TBGRD \
--min_capacity 1 \
--max_capacity 5 \
--query_nodes 5 \
--catalog_size 50 \
--demand_size 200 \
--min_weight 0 \
--max_weight 100 \
--time 1000 \
--action_selector_eta 0.1 \
--frequency -1 \
--trace_location traces/fixed_popularity_catalog_50.pkl
Executing the above script evaluates the caching policy TBGRD under the fixed popularity trace, with the following simulation parameters:
- select dtelekom network topology
- select TBGRG caching policy
- select from the set {1,...,5} the cache capacities
- generate demands from 5 nodes in the network
- set the catalog size to 50
- generate 200 different demands
- select from the interval [0,100] the edges weights
- set the simulation time to 100 seconds
- set the learning rate of the action selectorto 0.1
- freeze the colors selected by TBGRD
- load the fixed popularity trace
The synthetic traces are available in the traces/ directory, and we also provide a jupyter notebook tweak_traces.ipynb to tweak the traces.
| Fixed Popularity | Changing Popularity | Poisson Shot Noise |
|---|---|---|
 |
 |
 |
We provide the following plotting tools:
plot_sensitivity.pyis used for sensitivity analysis, and corresponding plots are insen_result/.- jupyer notebook
plot_bars_uc.ipynbis used to plot bar plots under different configurations and plot the update costs.
The directory res/ contains precomputed results, and figs/ directory contains precomputed figures.
The dependencies are specified in the requirements.txt file.
Simulate a Network of Caches
positional arguments:
outputfile Output file
optional arguments:
--max_capacity MAX_CAPACITY
Maximum capacity per cache (default: 6)
--min_capacity MIN_CAPACITY
Minimum capacity per cache (default: 3)
--max_weight MAX_WEIGHT
Maximum edge weight (default: 100.0)
--min_weight MIN_WEIGHT
Minimum edge weight (default: 1.0)
--max_rate MAX_RATE Maximum demand rate (default: 1.0)
--min_rate MIN_RATE Minimum demand rate (default: 1.0)
--time TIME Total simulation duration (default: 5000.0)
--warmup WARMUP Warmup time until measurements start (default: 0.1)
--catalog_size CATALOG_SIZE
Catalog size (default: 100)
--demand_size DEMAND_SIZE
Demand size (default: 500)
--demand_change_rate DEMAND_CHANGE_RATE
Demand change rate (default: 0.0)
--demand_distribution {powerlaw,uniform}
Demand distribution (default: powerlaw)
--powerlaw_exp POWERLAW_EXP
Power law exponent, used in demand distribution
(default: 1.2)
--query_nodes QUERY_NODES
Number of nodes generating queries (default: 10)
--graph_type {path,erdos_renyi,balanced_tree,hypercube,cicular_ladder,cycle,grid_2d,lollipop,expander,hypercube,star,barabasi_albert,watts_strogatz,regular,powerlaw_tree,small_world,geant,abilene,dtelekom,servicenetwork}
Graph type (default: erdos_renyi)
--graph_size GRAPH_SIZE
Network size (default: 100)
--graph_degree GRAPH_DEGREE
Degree. Used by balanced_tree, regular,
barabasi_albert, watts_strogatz (default: 4)
--graph_p GRAPH_P Probability, used in erdos_renyi, watts_strogatz
(default: 0.1)
--random_seed RANDOM_SEED
Random seed (default: 123456789)
--debug_level {INFO,DEBUG,WARNING,ERROR}
Debug Level (default: INFO)
--cache_type {LRU,FIFO,LFU,RR,EWMAGRAD,LMIN,TBGRD}
Networked Cache type (default: LRU)
--query_message_length QUERY_MESSAGE_LENGTH
Query message length (default: 0.0)
--response_message_length RESPONSE_MESSAGE_LENGTH
Response message length (default: 0.0)
--monitoring_rate MONITORING_RATE
Monitoring rate (default: 1.0)
--interpolate INTERPOLATE
Interpolate past states, used by LMIN (default: False)
--beta BETA Beta used in EWMA (default: 1.0)
--gamma GAMMA Gamma used in LMIN (default: 0.1)
--expon EXPON Exponent used in LMIN (default: 0.5)
--T T Shuffling period used in LMIN and TBGRD (default: 1)
--colors COLORS Number of colors used in TBGRD (default: 100)
--frequency FREQUENCY
Frequency of color updates used in TBGRD. When set to
-1 there is no update (default: -1)
--batch BATCH Whether requests are batched in TBGRD (default: False)
--samples SAMPLES Number of samples to estimate expected cache gain
(default: 100)
--trace_location TRACE_LOCATION
Generate demands from an external trace (default:
None)
--action_selector_eta ACTION_SELECTOR_ETA
Learning rate used in TBGRD ActionSelectors (default:
0.005)
--correlated_action_selectors CORRELATED_ACTION_SELECTORS
Enable correlated arms for action selectors (default:
False)
--adversarial_setting ADVERSARIAL_SETTING
Run abilene in adversarial setting (default: False)