Internet DRAFT - draft-c2f-ippm-big-data-alt-mark
draft-c2f-ippm-big-data-alt-mark
IPPM Working Group M. Cociglio
Internet-Draft Telecom Italia
Intended status: Experimental C. Corbo
Expires: May 3, 2021 Politecnico di Torino
G. Fioccola
Huawei Technologies
M. Nilo
Telecom Italia
R. Sisto
Politecnico di Torino
October 30, 2020
The Big Data Approach for Multipoint Alternate Marking method
draft-c2f-ippm-big-data-alt-mark-01
Abstract
This document introduces a new approach for the Alternate Marking
method. It is called Big Data Multipoint Alternate Marking method
and, starting from the methodology described in RFC 8321 and RFC
8889, it explains how to implement performance measurement analytics
on the Network Management System by analysing the raw data of the
network nodes.
Requirements Language
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119 [RFC2119].
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on May 3, 2021.
Cociglio, et al. Expires May 3, 2021 [Page 1]
Internet-Draft Big Data Multipoint AM October 2020
Copyright Notice
Copyright (c) 2020 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(https://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Marking Methods Classification . . . . . . . . . . . . . . . 3
3. Scenario and Background . . . . . . . . . . . . . . . . . . . 4
4. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.1. Data collecting . . . . . . . . . . . . . . . . . . . . . 6
4.2. Sending data . . . . . . . . . . . . . . . . . . . . . . 8
4.3. Preprocessing . . . . . . . . . . . . . . . . . . . . . . 8
4.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . 9
5. Security Considerations . . . . . . . . . . . . . . . . . . . 10
6. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 10
7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 10
8. References . . . . . . . . . . . . . . . . . . . . . . . . . 10
8.1. Normative References . . . . . . . . . . . . . . . . . . 10
8.2. Informative References . . . . . . . . . . . . . . . . . 11
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 11
1. Introduction
This document describes a scenario and a methodology that can be used
to get performance details from a monitored network. The approach is
inspired by the concepts illustrated in the Alternate Marking Method
(RFC 8321 [RFC8321]), Multipoint Alternate Marking Method (RFC 8889
[RFC8889]), and Hash Sampling (RFC 5474 [RFC5474] and RFC 5475
[RFC5475]).
In general the performance measurement results are based on a
posteriori calculation and the method is called Big Data Multipoint
Alternate Marking performance measurement.
Cociglio, et al. Expires May 3, 2021 [Page 2]
Internet-Draft Big Data Multipoint AM October 2020
The kinds of measurements are specified on the Network Management
System (NMS) and they can be split into two main categories: per
cluster and end-to-end.
o The per cluster approach includes all the details that refer to
each single cluster and provides a list of parameters that
characterize it (packet loss, mean delay).
o The end-to-end approach provides more general information about
the entire path (packet loss, mean delay).
The results can be provided on demand, in a non real-time processing
environment, and each one of them refers to a single monitoring
period, even if it is possible to broaden the search to more periods.
The basic mechanism of the Big data approach here introduced is the
Packet sampling. Packet sampling, which is performed through Hashing
Sampling technique (RFC 5474 [RFC5474] and RFC 5475 [RFC5475])
applied on all incoming traffic, without any flow distinction.
Nevertheless, thanks to data postprocessing, results are split by
flow afterwards, since the storage system memorizes the fields of the
packet headers that identify flows. The NMS, in fact, requires, as
input parameters, the flow identification fields as well as the
timestamp.
The use of hash sampling improves packet tracking performance and
thus overall performance. It allows to track the path followed by
each packet without further efforts by the NMS.
2. Marking Methods Classification
[I-D.mizrahi-ippm-compact-alternate-marking] presents a summary of
the alternate marking methods, and discusses the trade-offs among
them.
The methodologies are classified as follows:
o Double Marking,
o Single Marking,
o Hash-based Marking.
Double Marking and Single Marking are described in RFC 8321 [RFC8321]
and RFC 8889 [RFC8889].
Cociglio, et al. Expires May 3, 2021 [Page 3]
Internet-Draft Big Data Multipoint AM October 2020
While, Hash-based selection can be leveraged as a marking method,
allowing a zero-bit marking approach. As defined in RFC 5475
[RFC5475]:
A Hash Function h maps the Packet Content c, or some portion of
it, onto a Hash Range R. The packet is selected if h(c) is an
element of S, which is a subset of R called the Hash Selection
Range.
The Hash-Based marking requires the hash function and the set S to be
configured consistently across the measurement points. It is worth
mentioning that the duration between sampled packets depends only on
the hash value.
The single marking approach can be combined with hash-based sampling
as described in [I-D.mizrahi-ippm-compact-alternate-marking]: a
single marking bit is used for the loss measurement, while the hash-
based sampling is used to trigger delay measurement. In the same
way, the hash-based sampling can be used in multipoint network, and
this is explained in RFC 8889 [RFC8889].
3. Scenario and Background
The service provider's network is made up of a main backbone network
surrounded by routers that handle customers traffic input and output.
The proposed methodology requires that the traffic is marked before
entering the backbone network, by means of the Alternate Marking
technique. The marking process can be made by the edge routers or by
the customers itself, keeping in mind that it requires that the
markers are synchronized.
,..-..,_
.'` '.
+--+ ,' `. +--+
----|R1| / Backbone \ |R2|----
+--+/ Network \+--+
[unmarked | | [unmarked
traffic] | [marked | traffic]
+--+\ traffic] /+--+
----|R3| \ / |R4|----
+--+ `, ,' +--+
'. ,'`
`''-''``
Figure 1: Backbone Network
Cociglio, et al. Expires May 3, 2021 [Page 4]
Internet-Draft Big Data Multipoint AM October 2020
Only the marked traffic can be monitored. In case of one marking
bit, all the traffic must be marked. Instead, in case of two marking
bits, it is possible to mark the traffic partially, therefore the
results will not be affected by unmarked packets and will refer only
to the marked ones.
In order to apply the Alternate Marking methodology a time reference
period and a marking method must be fixed at the beginning. The time
reference period must consider the misalignment between the marking
source routers, clock error between network devices and the interval
we need to wait to avoid packets being out of order because of
network delay, as described in RFC 8321 [RFC8321] and RFC 8889
[RFC8889].
A possible marking method could use two bits of the header and set
them to 0x01, to identify a period, and to 0x10 to identify the next
one. This allows to distinguish between marked traffic and unmarked
traffic, instead of using just one bit, which can can generate
misunderstanding between the unmarked traffic (that has the marked
bit set to 0 by default) and the marked traffic (that alternates
between 0x0 and 0x1, with 0x0 as marker value and not as default).
As an alternative, it is possible to use just one marking bit and
utilize a filter based on IP subnets to exclude the flows from
monitoring. The flows that do not have to be monitored are those
internal to the network (that usually have private IP addresses).
To enable the Big Data approach for monitoring, the network nodes
require a packet collector, that is the agent installed on board of
the network node that collects measurements, based on the configured
Packet sampling criteria.
The portion of network to be monitored must be delimited by routers
with packet collector installed on. The rest of the network cannot
be monitored even if the traffic is marked. So, the size of the
monitored network depends on the network devices placement. However,
the size of the network surrounded by packet collectors must be less
than or equal to the size of the network with marked traffic.
It is worth highlighting that, if one marking bit is used, the
requirement is that all the ingress traffic from the boundary nodes
must be marked. While, if two marking bits are used, the marking is
applied even on flow basis by the boundary nodes delimiting the
monitored network and it is easy to recognize the marked traffic
within the network. Moreover, both in the case of one and two
marking bits, we need to ensure that all the marked traffic, both
ingress and egress, comes through the monitoring nodes in order to
guarantee to properly monitor the network.
Cociglio, et al. Expires May 3, 2021 [Page 5]
Internet-Draft Big Data Multipoint AM October 2020
4. Methodology
The method described here consists of the following steps:
1. Data collecting;
2. Sending data;
3. Preprocessing;
4. Results.
___________ _____________
| Packet | | |
|collector 1| -----> | |
|___________| |Preprocessing|
| |
| | ___________
___________ | | | |
| Packet | | (Grouping) | | Clusters |
|collector 2| ------> | | <-- |information|
|___________| | | | |
| | |___________|
| Results |
___________ | |
| Packet | | |
|collector 3| -------> | |
|___________| |_____________|
Figure 2: Outline of the Methodology
4.1. Data collecting
The Data collecting phase implies that, on board of the network
nodes, the packet collector analyses data passing through a network
interface. A packet collector needs to be placed into each router
interface we want to monitor.
The agent is configured by setting:
o the reference hash,
o the maximum number of packets to store,
o the alternate marking period duration,
Cociglio, et al. Expires May 3, 2021 [Page 6]
Internet-Draft Big Data Multipoint AM October 2020
o the one or the two alternate marking bits that identify the marked
flow,
o the interface to monitor,
o the flows to exclude from monitoring (i.e identified by header IP
fields): to be used only in case of one marking bit.
In general, if one alternate marking bit is available it is always
possible to identify the flow. In this case it must be used a filter
that excludes from monitoring the traffic flows that are not marked
in the network (e.g. IP addresses in use in the transit network that
often use private addresses) and, at the same time, it is needed to
make sure that all the ingress traffic is marked. This is a little
more complicated but helps to address the case of IPv4 where only one
bit could be available (i.e. so called unused bit). In this regard,
the flows to exclude from monitoring are needed only for the case of
one marking bit in order to identify the marked traffic. On the
other hand, these are not necessary in case of two marking bits
because it is already easy to identify the marked traffic and monitor
only this portion.
The Data collection is based on Hashing sampling described in RFC
5474 [RFC5474] and RFC 5475 [RFC5475].
The process is recursive. Each incoming packet is hashed, compared
with the reference hash, and recorded if the number of bits that are
matching are the same of those required by the packet collector .
When the number of matched packets exceed the maximum number
requested in configuration, the number of bits to match is increased
by one. At this point all the previous stored packets could be
potentially discarded and must be rechecked. So data are stored
temporally and are subject to changes, discards and additions. After
the period ends, previous data are still subject to change, but after
a guard band (reasonably L/2 if L is the period duration) data are
stored permanently and ready to be sent.
Note that the packet collector (carried out with probe) selects the
packets based on the configured parameters, so it works with every
incoming packet, without distinction. This greatly increases the
probe configuration easiness. Otherwise the probe should save all
possible flows (potentially too many), which would be too expensive
for the device, and need to be reconfigured if a new flow is
available for performance monitoring. On the other hand, this
increases the amount of data collected.
Stored data include two kind of details: one refers to each single
packet and the other one is about aggregate measures.
Cociglio, et al. Expires May 3, 2021 [Page 7]
Internet-Draft Big Data Multipoint AM October 2020
The first set of data includes the fields that identify the flow
(IP header fields), packet hash, timestamp when the packets come
in, period to which data refer.
The second set of data reports network interface identification,
total counted packets, total hashed packets, mean timestamp based
on all the timestamp of all packets that passed through the
interface, period.
4.2. Sending data
The Sending data phase is separated from the previous one. Once the
data has been stored and collected as logs by the network device
following the provisions of the theoretical model, the sending system
has only the task of carrying data safely and reliably. It is
possible to use a synchronous mechanism, in which the sending system
periodically checks the availability of new data, or an asynchronous
mechanism. In the last case when a new batch of data is ready, an
alert wakes up the sending system that carries them to the
destination.
4.3. Preprocessing
The Preprocessing phase has two main goals:
o aggregate input data to produce a new record that is ready to be
postprocessed and that makes it easier to obtain performance
parameters;
o decrease the total amount of data to store.
Although this step is not mandatory, it is recommended to speed up
subsequent operations and to give a better shape to the stored data
in order to fit well with the last queries.
Preprocessing can be done after data has been stored into the NMS in
an iterative loop that parses that periodically or just before to be
sent to NMS, through a consolidator, that collects data that comes
from all network devices, parses them and then sends them to the NMS.
However, in this phase it is possible to group incoming data from all
devices and determine the path followed by each sampled packet. In
order to do that, if the data are grouped by hash and ordered them by
timestamp, it is possible to outline the path.
After providing to the NMS the topology information and Clusters
partition of the monitored network, it is also possible to track the
crossed cluster for each couple of sorted data, by analyzing the
Cociglio, et al. Expires May 3, 2021 [Page 8]
Internet-Draft Big Data Multipoint AM October 2020
interface ID available in the stored record and comparing them with
the edge that characterizes the clusters available in the monitored
network.
4.4. Results
The Results phase involves the preprocessed records lay into
database. When necessary the storage system can be queried, in a
deferred time. The records are organized to fit well with the
queries that care about timing and loss aspects.
Results are achieved by querying the storage system properly.
Certainly, input parameters that identify which flow we are
addressing are required. Additionally, time reference is needed to
select only the packets of interest. The Big data system is aware of
flow identification fields and performs packet flow grouping on the
fly. The results described below can refer to different flows,
depending on which parameters have been specified for the query.
It is possible to deduce the cluster mean delay D_i (mean delay
referred to cluster i), by analyzing each record, computing delay d_j
(delay referred to record j) as difference between the two available
timestamps, that correspond to the input timestamp (when the packet
has gone into the cluster) and the output timestamp (when the packet
has gone out of that cluster), and summing it with all other delays;
then the result is divided by the number of records that refer to the
same cluster:
D_i = [d_0 + d_1 +...+ d_(N_i-1)] / N_i
Where D_i is the mean delay related to cluster i, d_j the delay
related to record j, N_i the total number of records belonging to
cluster i.
It could also be computed the end-to-end mean delay AD as the sum of
all delays available in our database, and dividing it by all the
records:
AD = [ad_0 + ad_1 +...+ ad_(M-1)] / M
Where AD is the end-to-end mean delay, ad_j the delay related to
records j, and M the total number of records.
If necessary, after observing an unusual cluster delay, it could be
possible to compute also max/avg/min link delay, by analyzing records
again, and exploiting the difference between the two timestamps.
Cociglio, et al. Expires May 3, 2021 [Page 9]
Internet-Draft Big Data Multipoint AM October 2020
Additionally, also details about loss are available. Since the total
packets are counted by each node, the sum of the input packets must
be equal to the sum of the output packets inside each cluster. If
their difference is greater than 0, then a loss has occurred, and the
result is the total loss. The total packet loss per cluster:
PL_i = [p_(i,0) + p_(i,1) +...+ p_(i,K-1)] - [p_(o,0) + p_(o,1) +...+
p_(o,L-1)]
Considering cluster i with K input nodes and L output nodes, the
calculation follows RFC 8889 [RFC8889].
In the same way it is possible to get the entire packet loss, as the
sum of all the packet loss per cluster. The same measure can be
obtained by using only the hashed packets, but in this case, we get
an approximate measurement that might reflect or not the real one.
Notice that all these measurements refers to the flow we specify as
input of the query and that the specified flow can include or not all
the sampled packets (e.g. filter on ip_src=0.0.0.0/0,
ip_dst=0.0.0.0/0, port_src=/, port_dst=/, type=tcp, outlines a flow
that includes all the TCP packets in an IP network).
5. Security Considerations
This document specifies a method of performing measurements that does
not directly affect Internet security or applications that run on the
Internet. However, implementation of this method must be mindful of
security and privacy concerns, as explained in RFC 8321 [RFC8321] and
RFC 8889 [RFC8889].
6. Acknowledgements
The authors would like to thank Guido Marchetto for the precious
contribution.
7. IANA Considerations
This document has no IANA actions
8. References
8.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
Cociglio, et al. Expires May 3, 2021 [Page 10]
Internet-Draft Big Data Multipoint AM October 2020
8.2. Informative References
[I-D.mizrahi-ippm-compact-alternate-marking]
Mizrahi, T., Arad, C., Fioccola, G., Cociglio, M., Chen,
M., Zheng, L., and G. Mirsky, "Compact Alternate Marking
Methods for Passive and Hybrid Performance Monitoring",
draft-mizrahi-ippm-compact-alternate-marking-05 (work in
progress), July 2019.
[RFC5474] Duffield, N., Ed., Chiou, D., Claise, B., Greenberg, A.,
Grossglauser, M., and J. Rexford, "A Framework for Packet
Selection and Reporting", RFC 5474, DOI 10.17487/RFC5474,
March 2009, <https://www.rfc-editor.org/info/rfc5474>.
[RFC5475] Zseby, T., Molina, M., Duffield, N., Niccolini, S., and F.
Raspall, "Sampling and Filtering Techniques for IP Packet
Selection", RFC 5475, DOI 10.17487/RFC5475, March 2009,
<https://www.rfc-editor.org/info/rfc5475>.
[RFC8321] Fioccola, G., Ed., Capello, A., Cociglio, M., Castaldelli,
L., Chen, M., Zheng, L., Mirsky, G., and T. Mizrahi,
"Alternate-Marking Method for Passive and Hybrid
Performance Monitoring", RFC 8321, DOI 10.17487/RFC8321,
January 2018, <https://www.rfc-editor.org/info/rfc8321>.
[RFC8889] Fioccola, G., Ed., Cociglio, M., Sapio, A., and R. Sisto,
"Multipoint Alternate-Marking Method for Passive and
Hybrid Performance Monitoring", RFC 8889,
DOI 10.17487/RFC8889, August 2020,
<https://www.rfc-editor.org/info/rfc8889>.
Authors' Addresses
Mauro Cociglio
Telecom Italia
Via Reiss Romoli, 274
Torino 10148
Italy
Email: mauro.cociglio@telecomitalia.it
Calogero Corbo
Politecnico di Torino
Email: corbocalo94@gmail.com
Cociglio, et al. Expires May 3, 2021 [Page 11]
Internet-Draft Big Data Multipoint AM October 2020
Giuseppe Fioccola
Huawei Technologies
Riesstrasse, 25
Munich 80992
Germany
Email: giuseppe.fioccola@huawei.com
Massimo Nilo
Telecom Italia
Via Reiss Romoli, 274
Torino 10148
Italy
Email: massimo.nilo@telecomitalia.it
Riccardo Sisto
Politecnico di Torino
Corso Duca degli Abruzzi, 24
Torino 10129
Italy
Email: riccardo.sisto@polito.it
Cociglio, et al. Expires May 3, 2021 [Page 12]
