Skip to content

Latest commit

 

History

History
73 lines (58 loc) · 2.69 KB

README.md

File metadata and controls

73 lines (58 loc) · 2.69 KB
Raw

Exercise 3: Investigate peer routing table sizes

The goal of this exercise is to build a distribution of per-peer routing table sizes in order to understand the difference between "partial" and "full-feed" peers. To do this, you should modify the script you wrote in Exercise 2 to count the number of unique prefixes per-peer in a RIB dump, then plot a distribution.

Steps

  1. Modify your script to process RIB (ribs) data instead of Updates (updates), select only a single collector (use route-views2 this time), and select a time interval that includes a RIB dump (e.g., 1484438400,1484442000).

  2. Modify your statistics collection to count the number of unique prefixes each peer announces (instead of number of updates).

  3. Use the code snippet below to generate a CDF of peer routing table sizes. You should see a "knee" at the right side of the graph that represents the full-feed peers. What would be a good threshold for determining if a peer is full-feed?

To minimize memory usage, you may assume that prefixes are not duplicated for a given peer and simply count them, rather than building a set of prefixes for each peer. Additionally, it takes my laptop around 5 minutes to process a RIB dump from route-views2, so you should add a short-circuit to your code during testing (e.g., break out of the loop after processing 500k elems).

CDF Plotting Code

To use this you'll need matplotlib and pandas. To install these, run:

sudo pip install matplotlib pandas

To count the prefixes per-peer, and plot a CDF:

import collections
import matplotlib.pyplot as plt
import pandas
import _pybgpstream

# !!TODO: add bgpstream init and filter code here

# helper function that simplifies extracting elems from a stream
def elem_generator(_stream, limit=None):
    _rec = _pybgpstream.BGPRecord()
    _cnt = 0
    while _stream.get_next_record(_rec):
        while True:
            _elem = _rec.get_next_elem()
            if _elem is None:
                break
            yield (_rec, _elem)
            _cnt += 1
            if limit is not None and _cnt == limit:
                return

# process the records and elems
stats = collections.defaultdict(int)  # stats[peer_id] = pfx_cnt
for (rec, elem) in elem_generator(bs):
    if "prefix" in elem.fields:
        peer_id = "|".join([rec.collector, str(elem.peer_asn),
                            elem.peer_address])
        stats[peer_id] += 1

df = pandas.DataFrame(sorted(stats.values()), columns=["pfx_cnt"])
df["cdf"] = df["pfx_cnt"].cumsum()/df["pfx_cnt"].sum()*100
df.plot(x="pfx_cnt", y="cdf")
plt.show()