pbhandar2 / pymimircache

cache analysis platform developed at Emory University and CMU

PyMimircache

NEWS

PyMimircache to appear at FAST tutorial.

PyMimircache is a cache trace analysis platform that supports

• comparison of different cache replacement algorithms

• visualization of cache traces

• easy plugging in your own cache replacement algorithm

Main users of PyMimircache includes researchers and system administrators, PyMimircache provides researchers an easy tool to study existing algorithms, devise and test new algorithms. While PyMimircache provides system administrators a simple tool helping them visualize and understand their cache.

PyMimircache is an independent Python3 platform that supports all the described features. Besides it also bundles with CMimircache for better performance. If you need a C/C++ platform, please check out CMimircache.

PyMimircache current supports algorithms include Least Recent Used(LRU), Least Frequent Used(LFU), Most Recent Used(MRU), First In First Out(FIFO), Segmented LRU(SLRU), Clock, Random, Optimal, Adaptive Replacement Cache(ARC). And we are actively adding more cache replacement algorithms.

Best of all is that you can easily and quickly implement your own cache replacement algorithm. See more information here

Dependency and Installation

System-wide library: glib, python3-pip, python3-matplotlib

On Ubuntu using the following command to install

jason@myMachine: ~$ sudo apt-get install libglib2.0-dev python3-pip python3-matplotlib

Python Dependency: numpy, scipy, matplotlib, heapdict, mmh3

jason@myMachine: ~$ sudo pip3 install heapdict mmh3

Installing PyMimircache

jason@myMachine: ~$ sudo pip3 install PyMimircache

Compatibility

PyMimircache only support Python3 and 64bit platform  

git clone

If you use Github repo, after git clone, do git submodules update --init to clone the CMimircache module.

Alternative using docker

As an alternative, you can use PyMimircache in a docker container, according to our simple benchmark, the performance difference between using a bare metal and a docker container is less than 10%.

Use interactive shell

To enter an interactive shell and do plotting, you can use

jason@myMachine: ~$ sudo docker run -it --rm -v $(pwd):/PyMimircache/scripts -v PATH/TO/DATA:/PyMimircache/data 1a1a11a/PyMimircache /bin/bash

After you run this command, you will be in a shell with everything ready, your current directory is mapped to /PyMimircache/scripts/ and your data directory is mapped to /PyMimircache/data. In addition, we have prepared a test dataset for you at /PyMimircache/testData.  

Run scripts directly

If you don't want to use an interactive shell and you have your script ready, then you can do

jason@myMachine: ~$ docker run --rm -v $(pwd):/PyMimircache/scripts -v PATH/TO/DATA:/PyMimircache/data 1a1a11a/PyMimircache python3 /PyMimircache/scripts/YOUR_PYTHON_SCRIPT.py

However, if you are new here or you have trouble using docker to run scripts directly, we suggest using interactive shell which can help you debug.

PyMimircache Tutorial

We have prepared a wonderful tutorial here. Check here for tutorial

PyMimircache Power

The power of PyMimircache

>>> from PyMimircache import Cachecow
>>> c = Cachecow()
>>> c.vscsi("trace.vscsi")      # find this data under data folder, other type of data supported too
>>> print(c.stat())
	# number of requests: 113872
	# number of uniq obj/blocks: 48974
	# cold miss ratio: 0.4301
	# top N popular (obj, num of requests):
	# [(3345071, 1630),
	#  (6160447, 1342),
	#  (6160455, 1341),
	#  (1313767, 652),
	#  (6160431, 360),
	#  (6160439, 360),
	#  (1313768, 326),
	#  (1329911, 326)]
	# number of obj/block accessed only once: 21049
	# frequency mean: 2.33
	# time span: 7200089885

>>> print(c.get_reuse_distance())
    # [-1 -1 -1 -1 -1 -1 11 7 11 8 8 8 -1 8]

>>> print(c.get_hit_ratio_dict("LRU", cache_size=20))
    # {0: 0.0, 1: 0.025256428270338627, 2: 0.031684698608964453, ... 20: 0.07794716875087819}

>>> c.plotHRCs(["LRU", "LFU", "Optimal"])

>>> c.heatmap('r', "hit_ratio_start_time_end_time", time_interval=10000000)

Hit Ratio Curve	Hit Ratio Heatmap

Next Major Releases

• Architecture: re-write PyMimircache pyGeneralProfiler (v0.3.0) [Done]
• Architecture: Separate mimircache into PyMimircache and CMimircache (v0.4.0) [Done]
• Architecture: Remove all CMimircache dependency in PyMimircache to allow CMimircache run independently (v0.5.0)
• Feature: Add block/obj size into cache replacement algorithm (v0.6.0)
• Feature: Add parallel support for LRUProfiling (v0.7.0)
• Feature: Add sampling layer for all profilers (v0.8.0)
• Milestone: Allow mimircache to interact with Memcached and Redis (v1.0.0)
• Feature: Add GPU profiling support (v1.2.0)
• Feature: Add Full Windows Support (v2.0.0)

Contributing

PyMimircache and CMimircache are created by Juncheng Yang of SimBioSys group at Emory University. CMimircache, previously Mimircache, was released as part of MITHRIL: Mining Sporadic Associations for Cache Prefetching. Juncheng Yang , Reza Karimi, Trausti Saemundsson, Avani Wildani, Ymir Vigfusson. ACM Symposium on Cloud Computing (SoCC), 2017.

This project has benefited from contributions from numerous people. You are more than welcome to make any contributions. Please create Pull Request for any changes.

LICENSE

PyMimircache is provided under GPLv3 license.

Related

libCacheSim: a high-performance C++ library for cache simulations