sylab/cacheus

The design and algorithms used in Cacheus are described in this USENIX FAST'21 paper and talk video: https://www.usenix.org/conference/fast21/presentation/rodriguez

cacheus's Issues

What is ikki.all.blkparse?

Hello, I want to know what ikki.all.blkparse is?

How to run correctly visual.py file

Hello, I'm looking it up visual.py file, the configuration file is found in example.config the "config ['graphs']" parameter is required. How to specify this parameter? Is there anything else we need to pay attention to in order to run it. in addition pollutionator.py What is the main purpose of the file. Thank!

Questions on trace data

Hello, I read your FAST21 paper and found this repo. I was trying to reproduce the experiments. May I ask if you can share the download script for the dataset? Actually, I did not find dataset with name "CloudPhysics", "CloudVPS" and "CloudCache". This prevents me from running your codebase. Thanks for your help!

I found the MSR Cambridge Traces in http://iotta.snia.org/historical_section?tracetype=block-io, i am not sure if this is the MSR dataset.

I found this link http://iotta.snia.org/traces/block-io/414 in a closed issue. But still I do not know where to find the "CloudPhysics", "CloudVPS" and "CloudCache" dataset.

I can run the cacheus algorithm on FIU/home1 dataset. I am trying to test it on more FIU sub-datasets. A good paper!

Questions about cacheus algorithm ?

Why choose lru algorithm from q queue?

The data that may match is in s queue.

[Question] LeCaR implementation

Hi, I have a question related to the LeCaR implementation in this repo.

    # Cache Miss
    def miss(self, oblock):
        evicted = None

        freq = 1
        if oblock in self.lru_hist:
            entry = self.lru_hist[oblock]
            freq = entry.freq + 1
            del self.lru_hist[oblock]
            reward_lru = -(self.discount_rate
                           **(self.time - entry.evicted_time))
            self.adjustWeights(reward_lru, 0)
        elif oblock in self.lfu_hist:
            entry = self.lfu_hist[oblock]
            freq = entry.freq + 1
            del self.lfu_hist[oblock]
            reward_lfu = -(self.discount_rate
                           **(self.time - entry.evicted_time))
            self.adjustWeights(0, reward_lfu)

        # If the cache is full, evict
        if len(self.lru) == self.cache_size:
            evicted, policy = self.evict()

        self.addToCache(oblock, freq)

        return evicted

In this case, the frequency of the newly-added object is the old frequency (before eviction) + 1. Why is this the case? The new object should have frequency of 1 considering that it has been evicted already by the cache, is that right?

On the problem of missing files

Hello, I'm a doctor of computer science and technology from Huazhong University of science and technology in China. After reading your team paper learning cache replacement with cacheus, we find the code according to the link, but after analysis, we find that the file seems to be missing. Specifically, there should also be a /lib folder in the /code/algs file directory, which is the basis of various cache replacement algorithms. However, it seems that there is no such file in the file you uploaded. I hope I can get an answer. Thanks

example.config

Demo doesn't work

HI
I am an undergraduate from Hust and I am very interested on your paper Cacheus.
However, after I downloaded Fiu trace and your source code, the demo doesn't work. When you run alecar6 experiment, you will find that the request function here is wrong and an TypeError will be raised for

place where the request is called_
for lba, write, _ in self.reader.read():
op, evicted = alg.request(lba, _ )

alecar6 request
def request(self, oblock):
miss = True
evicted = None

The number of the parameter here simply don't match with each other here.
It seems that the parameter ts here is missing.

About the min algorithm

What is the principle of Min algorithm? Is there any relevant literature？ I can't understand the meaning of your code. Thank you very much!!