Internet DRAFT - draft-hong-icnrg-bloomfilterbased-name-resolution
draft-hong-icnrg-bloomfilterbased-name-resolution
ICNRG J. Hong
Internet Draft ETRI
Intended status: Informational W. Chun
Expires: April 2016 HUFS
H. Jung
ETRI
October 19, 2015
Bloom Filter-based Flat Name Resolution System for ICN
draft-hong-icnrg-bloomfilterbased-name-resolution-05.txt
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
This document may contain material from IETF Documents or IETF
Contributions published or made publicly available before November
10, 2008. The person(s) controlling the copyright in some of this
material may not have granted the IETF Trust the right to allow
modifications of such material outside the IETF Standards Process.
Without obtaining an adequate license from the person(s) controlling
the copyright in such materials, this document may not be modified
outside the IETF Standards Process, and derivative works of it may
not be created outside the IETF Standards Process, except to format
it for publication as an RFC or to translate it into languages other
than English.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
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."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html
Hong Expires April 19
, 2016 [Page 1]
Internet-Draft Bloom filter-based NRS October 2015
This Internet-Draft will expire on April 19, 2016.
Copyright Notice
Copyright (c) 2015 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
(http://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.
Abstract
In information-centric networking (ICN), uniquely identifiable and
location independent names are assigned directly to the named data
which raises scalability issues and they get even worse with flat
names. Accordingly, name resolution system required for lookup-by-
name routing in ICN has to be designed to scale, also considering
mobility support. In this draft, a bloom filter-based flat name
resolution system (B-NRS) is proposed where the bloom filter as an
aggregated form of names and hierarchical structure of the B-NRS are
exploited to address the scalability issues.
Table of Contents
1. Introduction ................................................ 4
2. NRS Requirements ............................................ 5
2.1. Scalability ............................................ 5
2.2. Fast resolution......................................... 5
2.3. Fast update ............................................ 6
2.4. Resilience ............................................. 6
2.5. Security ............................................... 6
3. Bloom Filter-based Flat Name Resolution System (B-NRS) ........ 6
3.1. System structure ........................................ 6
3.2. B-NRS Server Components ................................. 8
3.2.1. Name Lookup Table .................................. 8
3.2.2. Bloom Filter ....................................... 8
3.3. Key Operations ......................................... 9
3.3.1. Name Registration .................................. 9
3.3.2. Locator Update .................................... 10
3.3.3. Locator Lookup .................................... 11
Hong Expires April 19, 2016 [Page 2]
Internet-Draft Bloom filter-based NRS October 2015
4. Comparison of B-NRS with Other NRSs ......................... 12
5. Implementation Issues ....................................... 14
5.1. False Positive ........................................ 14
5.2. Membership Deletion .................................... 15
5.2.1. Use case ......................................... 16
6. Implementation of B-NRS ..................................... 16
6.1. Protocol Message ....................................... 17
7. Security Considerations ..................................... 19
8. IANA Considerations ........................................ 19
9. References ................................................. 20
9.1. Normative References ................................... 20
9.2. Informative References ................................. 20
A.1. Authors' Addresses ..................................... 22
Hong Expires April 19, 2016 [Page 3]
Internet-Draft Bloom filter-based NRS October 2015
1. Introduction
In contrast to the host-centric networking in the current Internet,
the primary communication object in information-centric networking
(ICN) is named data, where uniquely identifiable and location
independent name is assigned directly to the named data. This shift
raises scalability issues to a new level. The current Internet is
addressing on the order of 10^9 nodes, whereas the number of
addressable ICN objects is expected to be several orders of
magnitude higher [ICNRG charter]. Accordingly, name resolution
system required both for lookup-by-name routing in ICN [ICN
Challenges] and for ICN-IoT architecture [ICN-IoT] has to be
designed to scale, also considering mobility support.
In this draft, we propose a bloom filter-based flat name resolution
system (B-NRS) which maintains and resolves the binding between
names and locators, i.e. B-NRS takes a name as its input and
produces the locator sets that the name is currently associated
with. We assume that the locator independent names are flat since
the flat names provide some advantages compared to hierarchical
ones, such as higher flexibility, simpler name allocation and
benefits in terms of persistency and privacy [Ghodsi, ITU]. On the
other hand, scalability becomes the most important challenge on
designing the NRS supporting flat names. It is because of the ever
increasing number of names in the network and no possible way to
compactly represent the flat names such as the aggregation in IP
addresses.
In order to address the scalability issue in designing the NRS for
flat name, we need to aggregate names in any shape of type. One
popular technique for flat name is Distributed Hashing Table (DHT)
based approach [Hanka, Luo, Ahlgren, Mathy], where multiple servers
form circular linked list and the bindings are stored in the
appropriate server. However, the DHT technique has some drawbacks;
the binding must be stored in a server other than the owner's, which
causes a serious trust problem related to the authority issue and
lookup message may be propagated through the long paths.
In this draft, to overcome the drawbacks of DHT, we exploit the
bloom filter as an aggregated form of names and hierarchically
construct the B-NRS. One of the major benefits of the bloom filter
is a fixed constant time of insertion and search which is completely
independent of the number of names already in the set. Another
important and powerful property of bloom filter is the efficient
support for union of bloom filters with the same size and set of
hash functions which can be implemented with bitwise OR. However,
bloom filter also has some drawbacks; false positive and no member
Hong Expires April 19, 2016 [Page 4]
Internet-Draft Bloom filter-based NRS October 2015
deletion. Although there is no way to get rid of the false positive,
it can be minimized by choosing the right parameters. The deletion
problem is also taken care by periodic reconstruct of the bloom
filters or by using variants of the bloom filter such as the
counting bloom filter.
We note that the B-NRS in this draft does not require any specific
mechanism for registering names, since names have no structure and
can be registered to any B-NRS server with no constraint. Thus, the
B-NRS needs only lookup mechanism. Whereas in the DHT-based system,
the lookup message for a name is forwarded by the same way how to
register the name.
2. NRS Requirements
Name resolution system (NRS) may become the bottleneck of the
network when the signaling overhead of the location update and
lookup becomes very large. Thus, the NRS must provide fast update
and lookup for good performance since its basic functionality is to
return the current locator for a given name. The NRS also must be
secure and resilient because there is no way to respond to the
querying message if the NRS is attacked. Obviously, the NRS must be
scalable to the number of the ever-increasing ICN objects, i.e.
names. Therefore, in this section, we discuss such requirements of
the NRS.
2.1. Scalability
In ICN, the primary communication object is named data, where
uniquely identifiable and location independent name is assigned
directly to the named data. This raises scalability issues to a new
level. The current Internet is addressing even on the order of 10^9
nodes, whereas the number of addressable ICN objects is expected to
be several orders of magnitude higher considering sensor data,
vehicular, Internet of things, etc. Accordingly, the NRS should be
able to fully cover the ever-increasing number of ICN objects.
2.2. Fast resolution
A fundamental problem with any global query server network is that
the requestor who sends the name resolving request may significantly
delay or drop the initial packet of a new session if the resolution
time gets too long. Thus, the resolution time should be sufficiently
low so it does not affect much the overall system performance.
Hong Expires April 19, 2016 [Page 5]
Internet-Draft Bloom filter-based NRS October 2015
2.3. Fast update
When a named date moves and changes its point of attachment to
Internet or a multi-homed device shuts down one of its physical
interface, it needs to update the old information with the new one
or delete the deprecated information in NRS. Thus, the NRS should
adapt quickly with such changes.
2.4. Resilience
If the NRS fails, there is mostly no way for the requestor to reach
other end information since the requester knows only its names.
Therefore, the NRS must not fail.
2.5. Security
The NRS can be a potential target for attacks such as denial-of-
service attacks. These types of attacks are difficult to prevent.
Thus, updates to the NRS or responses from NRS server should be
authenticated.
3. Bloom Filter-based Flat Name Resolution System (B-NRS)
We propose a bloom filter-based name resolution system (B-NRS) for
supporting flat name which maintains and resolves the binding
between names and locators.
3.1. System structure
We construct the B-NRS hierarchically by defining a network of B-NRS
servers, which consists of a forest by several disjoint trees. The
network of B-NRS servers is defined by both parent-child and peering
relationships.
Figure 1 is an example of the B-NRS structure which consists of 8 B-
NRS servers forming a tree, where there exists the peering
relationship between S2 and S3. The peering relationship is allowed
for better performance by reducing the overhead for the B-NRS at the
top of the tree. A leaf B-NRS server knows every single name/locator
pair that it manages but nothing else. The intermediate B-NRS
servers know the name/locator pair for all names that are directly
registered to them and also possess only information about the names
that their descendant and peer B-NRS servers manage. Although there
is a single tree in figure 1, if we assume there are several trees
forming a forest, then the B-NRS servers are fully peered at the top
Hong Expires April 19, 2016 [Page 6]
Internet-Draft Bloom filter-based NRS October 2015
of the trees. This means that each server shares its knowledge of
all names that it manages with its peers.
We note that we have been very careful in distinguishing between the
name/locator pair information and the name information. This
distinction is necessary to provide a different level of information
abstraction, which is naturally achieved through the hierarchical B-
NRS structure and the use of bloom filters.
+----+
| S1 |
+----+
/ \
/ \
/ \
/ \
/ \
/ \
/ \
+----+ +----+
| S2 |*******************| S3 |
+----+ +----+
/ | \ /\
/ | \ / \
/ | \ / \
/ | \ / \
/ | \ / \
/ | \ / \
/ | \ / \
+----+ +----+ +----+ +----+ +----+
| S4 | | S5 | | S6 | | S7 | | S8 |
+----+ +----+ +----+ +----+ +----+
Legend:
+---+
| S | B-NRS Server
+---+
----- Parent-child relationship
***** Peering relationship
Figure 1. An example of B-NRS structure
Hong Expires April 19, 2016 [Page 7]
Internet-Draft Bloom filter-based NRS October 2015
3.2. B-NRS Server Components
A B-NRS server consists of a name lookup table and multiple bloom
filters.
3.2.1. Name Lookup Table
Name lookup table stores the binding between names and locators for
all names which are directly registered to the BRS server. The
associated locator for a certain name can be more than one. So, the
locator information is stored as a set shown in table 1. Name lookup
table takes a name as the input and produces its associated locator
sets as the output.
Table 1. Lookup table
=================================
Name | Locators
=================================
N1 | LOC1
N2 | LOC2-1, LOC2-2
N3 | -
N4 | LOC4-1, LOC4-2, LOC4-3
=================================
3.2.2. Bloom Filter
We utilize bloom filters as an aggregated form of names at each B-
NRS server. B-NRS servers announce their name set to the other B-NRS
servers. Instead of announcing the whole list of names, bloom filter
as an aggregated form of names is announced. When announcing its
name set to its peers or parents, the B-NRS server announces the
union of name sets of all child B-NRS servers. Union of child name
sets can be built by using the characteristic of bloom filer that
bloom filter for union of sets can be built merely by bitwise 'OR'
operation on all the sets.
Thus, each B-NRS server stores bloom filters for itself, from
children, and from peers depicted in figure 2. The B-NRS server
stores n+m+1 bloom filters in figure 2, where n is the number of
child B-NRS servers and m is the number of peer B-NRS servers.
We note that the forest of B-NRS servers retains the loop-free
property for the use of bloom filter.
Hong Expires April 19, 2016 [Page 8]
Internet-Draft Bloom filter-based NRS October 2015
/ ------------------------- \
/ | BF for its own | \
/ -------------------------- \ Bitwise OR
+----------------+ / -------------------------- / To Parents and Peers
| B-NRS Server | | BFs from Child 1 to n | /
+----------------+ \ -------------------------- /
\ --------------------------
\ | BFs from Peer 1 to m |
\ ---------------------------
Figure 2. B-NRS server components
3.3. Key Operations
3.3.1. Name Registration
When a communication entity attempts to join the network, it must
register itself in at least one B-NRS server. In this draft, it is
allowed that the communication entity can be registered in any
arbitrary B-NRS server since names have no structure.
Upon receiving the registration request from the communication
entity, the B-NRS server registers the name to its lookup table. The
locators for the name are stored in the table when the communication
entity for the name is actually present into the network. We
separate this as the operation of locator update from the name
registration.
The name registration is along with bloom filter update. When a
communication entity is registered in a B-NRS server, the
registration information is extracted from its name using the hash
functions for its bloom filter and inserted into its own bloom
filter first and then the B-NRS server updates bloom filters for its
parents and peers, where this recursion holds until bloom filters at
the top of trees are completely updated.
Figure 3 shows an example of the name registration and bloom filter
updates, where a new name is registered at the B-NRS server, S4. It
inserts information of the new name first into its own bloom filter
and updates its parent, S2. Then, S2 updates its parent, S1 and its
peer, S3.
Hong Expires April 19, 2016 [Page 9]
Internet-Draft Bloom filter-based NRS October 2015
When names are deleted from the lookup table, we need to adopt a
certain mechanism to update the bloom filters for the deletion since
bloom filter cannot handle the deletion by itself. Thus, we use the
periodic refresh technique that bloom filters with registered names
are rebuilt periodically and followed by bloom filter updates.
(3)BF
Update +----+
-------->| S1 |
| +----+
| / \
| / \
| / \
| / \
| / \
| / \
(2)BF | / \
Update +----+ (3)BF Update +----+
----> | S2 |------------------>| S3 |
| +----+*******************+----+
| / | \ /\
| / | \ / \
| / | \ / \
| / | \ / \
| / | \ / \
| / | \ / \
|/ | \ / \
+----+ +----+ +----+ +----+ +----+
| S4 | | S5 | | S6 | | S7 | | S8 |
+----+ +----+ +----+ +----+ +----+
^^
||
|| (1)Name registration
||
Figure 3. Name registration and BF update
3.3.2. Locator Update
When a communication entity actually presents in the network, the
locator update is occurred, where the gateway sends the locator
update message to the correspondent B-NRS server and the locator
associated with the name is stored in the lookup table. If the name
Hong Expires April 19, 2016 [Page 10]
Internet-Draft Bloom filter-based NRS October 2015
has multiple locators, then they are stored as a set of locators for
the name. Through the bloom filter test of the name, the locator
update messages are forwarded into the lookup table where the name
is actually stored.
When the communication entity depresents from the network, the
locators for the name is deleted from the lookup table by the
locator update message as well. Table 1 shows the depresence of
entity for the name, N3. We note that changing locators has no
effect on the structure of the B-NRS and mobility is easily
supported.
3.3.3. Locator Lookup
The lookup operation is to find the locator information for a given
name. The simplest case is when the source object tries to
communicate with the destination object registered in the same B-NRS
server. B-NRS server always searches for the destination name in its
own lookup table first so the locator information is acquired at the
first lookup in such a case.
A harder, but more interesting, case is when the destination object
is registered in the other B-NRS server with the source object. In
this case, the B-NRS server would quickly learn that the destination
object is not registered in the same B-NRS server by a simple search
of its lookup table. Then, it searches bloom filters for its child
and peer B-NRS servers. If none of the bloom filters return a
positive answer, the lookup request message is forwarded to its
parent B-NRS server. On the other hand, if any of bloom filters
return a positive answer, the lookup request message is forwarded to
every B-NRS server that corresponds to the bloom filters with
positive answers. We note that because of the false positives of the
bloom filter, multiple bloom filters may return positive answers.
This search is done recursively, and the locator information for the
destination name can eventually be found. Once the locator
information is found, it is delivered to the source object by the
lookup reply message which takes the reverse path of the lookup
request message.
Figure 4 is an example of lookup and registration processes where
the lookup message for a name which is registered at S8 is received
by S4. Then, the lookup message is forwarded to S2. Since S2 is
peered with S3, S2 forwards it to S3 not to S1. S3 forwards it to
Hong Expires April 19, 2016 [Page 11]
Internet-Draft Bloom filter-based NRS October 2015
S8. The reply message takes the reverse path of the lookup request
message, i.e., S8->S3->S2->S4.
+----+
| S1 |
+----+
/ \
/ \
/ \
/ \
/ \
/ \
/ \
(2)Lookup +----+ (3)Lookup +----+ (4)Lookup
----> | S2 |<----------------->| S3 |<------
| +----+*******************+----+ |
| / | \ (6)Reply /\ |
| / | \ / \ |
(7)Reply | / | \ / \ |(5)Reply
| / | \ / \ |
| / | \ / \ |
| / | \ / \ |
v/ | \ / \ v
(1)Lookup +----+ +----+ +----+ +----+ +----+
<-------->| S4 | | S5 | | S6 | | S7 | | S8 |
(8)Reply +----+ +----+ +----+ +----+ +----+
Figure 4. Lookup and reply
4. Comparison of B-NRS with Other NRSs
One of the critical challenges in designing NRS is scalability due
to the ever increasing number of names. In order to overcome this
issue, names need to be distributed and also aggregated in any shape
of type especially for flat names. One popular technique to
distribute and aggregate names is to use DHT (Distributed Hash
Table). However, DHT has several drawbacks such as ownership,
deployment, locality, etc. Thus, we exploit the bloom filter as an
aggregated form of names and hierarchically construct the NRS.
As illustrated in figure 5, NRS can be roughly divided into two
types: centralized vs. distributed. Then, the distributed type can
be divided again into two approaches: DHT-based vs. all else. DMap
(Direct Mapping) [DMap] and MDHT (Multiple DHT) [MDHT] are examples
of DHT-based approach. DMap is proposed by MF (MobilityFirst) which
Hong Expires April 19, 2016 [Page 12]
Internet-Draft Bloom filter-based NRS October 2015
is one of the Future Internet architecture projects funded by NSF in
US and MDHT is by SAIL (Scalable and Adaptive Internet Solutions)
which is an EU-funded project. B-NRS belongs to the distributed type
but not DHT-based approach.
*==================== NRS ===================*
| |
| *========================* |
| * Distributed * Centralized |
| * * |
| * *==========* * |
| * * DHT-based * * |
| * * o MF-DMap * o B-NRS * |
| * * o SAIL-MDHT * * |
| * *===========* * |
| * * |
| *=======================* |
| |
*=============================================*
Figure 5. A simple Venn diagram categorizing NRS
Table 2 presents the comparison of B-NRS with DMap and MDHT in
respect of scalability, lookup latency and locator update.
For scalability, we compare how many names can be scalable for each
NRS. DMap assumes that the number of names is 5*10^9, whereas MDHT
and B-NRS assume that it is 10^15.
We define the lookup latency as the multiple of the number of hops,
H, and the processing time per hop, T(N), which is proportional to
the number of table entries, N. The lookup latencies for both DMap
and MDHT are increasing proportionally to the number of hops and the
number of table entries at each hop since the table lookup is
processed at each hop. However, the lookup latency for B-NRS is
dependent only to the number of hops since BF takes a fixed constant
time, C, for searching. Even though each B-NRS server has several
bloom filters, they are independent to each other and can be
parallelized in a hardware implementation.
For locator update, we look at the staleness. Both DMap and MDHT do
the location update periodically so the staleness occurs during it
is not updated. However, the staleness for B-NRS occurs with
probability 0 since it does the location update in real time.
Hong Expires April 19, 2016 [Page 13]
Internet-Draft Bloom filter-based NRS October 2015
Table 2. Comparison of B-NRS with DMap and MDHT (N and H are the
number of table entries and hops, respectively and C is a constant.)
====================================================================
Design goal| Scalability | Lookup latency | Locator update
====================================================================
Metric | number of | number of hops | Staleness
| names | * processing |
| | time per hop |
====================================================================
MF-DMap | ~5*10^9 | H*T(N) | periodic update:
| | | occur
--------------------------------------------------------------------
SAIL-MDHT | ~10^15 | H*T(N) | periodic update:
| | | occur
--------------------------------------------------------------------
B-NRS | ~10^15 | H*C | real time update:
| | | occur with
| | | probability 0
====================================================================
5. Implementation Issues
Bloom filter has the well-known drawbacks such as false positive and
no membership deletion. However, the false positive can be minimized
by choosing the right parameters and the deletion problem can also
be taken care by adopting a certain mechanism to update the bloom
filters for the deletion such as the counting bloom filter, periodic
reconstruct of bloom filter, etc.
5.1. False Positive
The width of a bloom filter is directly related to the false
positive rate for fixed number of hash functions, the length of the
bloom filter is inversely proportional to the false positive rate.
Although a lengthier bloom filter is ideal for minimizing the false
positive rate but increasing the B-NRS search efficiency, it creates
a burden when filter information are exchanged among B-NRS servers.
In addition, since a leaf B-NRS server has a smaller number of names
that it needs to manage, it makes sense to use a smaller bloom
filter than the B-NRS servers at the higher level of the B-NRS
hierarchy. However, the variable bloom filter length approach must
Hong Expires April 19, 2016 [Page 14]
Internet-Draft Bloom filter-based NRS October 2015
be done with care since the key property, union of bloom filter via
bit-wise AND operation, may be lost when variable length bloom
filters are used.
5.2. Membership Deletion
One of the main advantages of the bloom filter is that data
insertion and search can be done in a constant time. However, its
major drawback is that a bloom filter does not have an efficient
method of supporting data deletion. Of course, there are variants of
the bloom filter to overcome the deletion issue. For example, the
counting bloom filter supports the deletion by associating a counter
to every bit of the bloom filter, where data insertion corresponds
to incrementing the counters associated with the bits; data deletion
to decrementing the counters; and query to checking whether the
counters are positive. However, since each counter needs to have
sufficient number of bits to prevent overflow; thus, it is a less
space efficient than the traditional bloom filter. The space
efficiency is critical to our B-NRS since bloom filters are
exchanged among B-NRS servers and it is directly proportional to the
size of exchanged control messages.
Because of this drawback of deletion of bloom filter, B-NRS needs to
be carefully designed to support dynamic registration and
deregistration of communicating entity.
In one extreme case, even if the de-registration were to be
completely ignored by the B-NRS, the B-NRS would eventually be able
to find the locator for a given name. This method will generate the
fewest number of control messages (bloom filter updates) but the
query would become inefficient since this would significantly
increase the false positive rates.
The other extreme case would be to update the entire B-NRS whenever
there is a single de-registration. Although this method would have
the lowest false positive rates, and thus, would have the lowest
average number of queries to find the name/locator pair, it would
have a very high control message load since there would be a lot of
bloom filter exchanges among B-NRS servers.
Certainly, the B-NRS will operate within these two extreme bounds,
and the optimal rate is a design parameter in building the B-NRS
system.
Hong Expires April 19, 2016 [Page 15]
Internet-Draft Bloom filter-based NRS October 2015
B-NRS overcome the deletion issue by periodically rebuilding bloom
filters using the shadow memory, so called periodic refresh. The
refresh frequency can be a day, a week, a month, etc. When B-NRS is
refreshed, names in a name lookup table are inserted into the new
bloom filter at a time and the merged bloom filters by bitwise OR
are announced to parent and peer B-NRS servers. For better
performance, the lossless compressed bloom filter can be used to
announce the merged bloom filter. We note that the false positive
probability certainly increases until all bloom filters are replaced
by new bloom filters.
5.2.1. Use case
What happened if the deleted name is requested before the bloom
filters are refreshed? The lookup message for the deleted name will
be forwarded to the B-NRS server which stored the name. Once it gets
the server, it will learn that the name does not exist in the lookup
table of the server. Then, the lookup message is processed as a
false positive case so that it would eventually return a response
that there is no such name registered in the system. Therefore, the
requestor would get a correct corresponding response even when the
bloom filters are not refreshed.
Now, what is the difference between before and after the bloom
filter refreshment? The requestor for the deleted name will get the
same response in both case but the response will be processed much
sooner after the refreshment since the lookup will not be forwarded
to the server which stored the name.
As a result, it may not be fatal in B-NRS that bloom filter cannot
handle the membership deletion. However, the periodic refreshment of
bloom filters are necessary for the better performance and
management.
6. Implementation of B-NRS
We have created prototypes for B-NRS: NRS server, top server, and
client. Although all B-NRS servers perform the same functions, we
separate top server from the others for convenient implementation.
We have utilized the parallel process of a graphics processor unit
(GPU) to accelerate the performance of BF check at each B-NRS server
resulting in low latency.
Hong Expires April 19, 2016 [Page 16]
Internet-Draft Bloom filter-based NRS October 2015
We have used an algorithm for the GPU usage. The main idea of the
algorithm is to enable to extract only the corresponding bits for
the given name check from all BFs at each server to GPU memory and
check the extracted bits in parallel to see if any chunk gives 1 by
bitwise 'AND' operation. In this implementation, we use 16Mb BF size
and 11 hash functions to keep the false positive probability less 6 information at a maximum of 10 names. We have used the static tree
structure of B-NRS which is managed by configuration files of each
server. We have also implemented the B-NRG by using CPUs to see the
effect of the GPU usage on performance. It showed that the search
time of a number of bloom filters with GPU was almost constant up to
the number of GPU cores. In other words, as expected, the search
time with CPU was linearly increasing according to the number of
bloom filters. The search time with GPU became shorter than the time
with CPU when the number of bloom filters was greater than a certain
amount, which value is dependent to the specification of GPU and
CPU. Using GPU is also much more cost-effective compared to CPU.
This results are powerful when the number of bloom filters in a B-
NRS server is huge. A number of bloom filters in a B-NRS server
means that the server has the amount of child servers including
peers. Having a number of child server is desirable because it is
the way to reduce the height of the B-NRS hierarchy resulting in
reducing the number of B-NRS server accesses per a lookup.
6.1. Protocol Message
We keep the flat name size as 24 bytes and use the UDP communication
with port number, 7979 in the implementation. Prot in protocol
messages is the protocol type of 1 byte size. Locator is defined as
a variable length string.
O Name registration
+----------------------------------------+
| Prot | Name |
+----------------------------------------+
Hong Expires April 19, 2016 [Page 17]
Internet-Draft Bloom filter-based NRS October 2015
O Locator update
Locator update message is divided into three types: Add, Delete, and
Replace.
+---------------------------------------------------------+
| Prot | Mode | Name | Locator length | Locator |
+---------------------------------------------------------+
Mode is the type of locator update.
O Locator lookup
+----------------------------------------+
| Prot | Name |
+----------------------------------------+
O Name deregistration
It deletes the name and the corresponding locators from name lookup
table.
+----------------------------------------+
| Prot | Name |
+----------------------------------------+
O BF update
+----------------------------------------+
| Prot | Name |
+----------------------------------------+
Hong Expires April 19, 2016 [Page 18]
Internet-Draft Bloom filter-based NRS October 2015
O CMD_Lookup
It is the locator lookup message between B-NRS servers.
+-----------------------------------------------+
| Prot | Name | Client IP | Up/Down | Depth |
+-----------------------------------------------+
It keeps the IP address of the client who creates the locator lookup
message so the locator information could be delivered directly to
the client once it is found. Up denotes that the lookup message is
to parent server and Down is to child servers. We increase the Depth
by 1 whenever the message is forwarded to child. We keep the depth
information because of the false positive of BF.
O CMD_Lookup NACK
When BF check fails, it is sent to parent server.
+-----------------------------------------------+
| Prot | Name | Client IP | Up/Down | Depth |
+-----------------------------------------------+
7. Security Considerations
False positive error is one of the well-known drawbacks of bloom
filter and there is no way to get rid of it. Thus, it can be an
attack point. For example, if an attacker puts wrong information
into bloom filters of B-NRS in order to increase the false positive
error rate resulting in getting traffics to go far away and
consuming resource, then the performance degradation may occur until
the B-NRS is refreshed. Once B-NRS is rebuilt, there will be only
probabilistic false positive error rate not the deterministic one.
8. IANA Considerations
TBD
Hong Expires April 19, 2016 [Page 19]
Internet-Draft Bloom filter-based NRS October 2015
9. References
9.1. Normative References
9.2. Informative References
[ICNRG charter] http://irtf.org/icnrg
[ICN Challenges] D.Kutscher, S. Eum, K. Pentikousis, I. Psaras, D.
Corujo, D. Saucez, T. Schmidt, and M. Waehlisch, "ICN
Research Challenges ", draft-kutscher-icnrg-challenges-02,
February 2014.
[ICN-IoT] Y. Zhang, D. Raychadhuri, L. Grieco, E. Baccelli, J.
Burke, R. Ravindran, and G. Wang, "ICN based Architecture
for IoT -Requirements and challenges", draft-zhang-iot-
icn-challenges-01, December 2014.
[Ghodsi] A. Ghodsi, T. Koponen, J. Rajahalme, P. Sarolahti, and
Shenker, "Naming in Content-Oriented Architectures," In
Proceedings of the SIGCOMM ICN'11, August 19, 2011,
Toronto, Ontario, Canada.
[ITU] International Telecommunication Union (ITU), "ITU-T
Recommendation Y.3031 - Identification framework in future
networks," available at: http://www.itu.int/rec/T-REC-
Y.3031-201205-P/en, 2012.
[Hanka] O. Hanka, C. Spleiss, G. Kunzmann, and J. Eberspacher, "A
novel DHTbased network architecture for the next
generation internet," Eighth International Conference on
Networks, Cancun, Mexico, March 2009.
[Luo] H. Luo, Y. Qin, and H. Zhang, "A DHT-Based Identifier-to-
Locator Mapping Scheme for a Scalable Internet," IEEE
Transactions on Parallel and Distributed Systems, October
2009.
[Ahlgren] B. Ahlgren, J. Arkko, L. Eggert, and J. Rajahalme, "A node
identity internetworking architecture," in INFOCOM 2006.
25th IEEE International Conference on Computer
Communications Proceedings. Washington, DC, USA: IEEE
Computer Society, April 2006, pp. 1-6.
Hong Expires April 19, 2016 [Page 20]
Internet-Draft Bloom filter-based NRS October 2015
[Mathy] L. Mathy and L. Iannone, "LISP-DHT: Towards a DHT to map
identifiers onto locators," in ReArch'08. Madrid, Spain:
ACM, December 2008.
[Fab1999] Faber, T., Touch, J. and W. Yue, "The TIME-WAIT state in
TCP and Its Effect on Busy Servers", Proc. Infocom 1999
pp. 1573-1583.
[DMap] T. Vu, A. Baid, Y. Zhang, T. Nguyeny, J. Fukuyamaz, R.
Martin, and D. Raychaudhuri, "DMap: A Shared Hosting
Scheme for Dynamic Identifier to Locator Mappings in the
Global Internet," Proceedings of the IEEE International
Conference on Distributed Computing Systems, pp. 698-707,
2012.
[MDHT] M. D'Ambrosio, C. Dannewitz, H. Karl, and V. Vercellone,
"MDHT: A Hierarchical Name Resolution Service for
Information-centric Networks," ICN'11, August 19, 2011,
Toronto, Ontario, Canada.
Hong Expires April 19, 2016 [Page 21]
Internet-Draft Bloom filter-based NRS October 2015
A.1. Authors' Addresses
Jungha Hong
ETRI
218 Gajeong-ro, Yuseong-gu, Daejeon, Korea
Email: jhong@etri.re.kr
Woojik Chun
Hankuk University of Foreign Strudies
81, Oedae-ro, Mohyeon-myeon, Cheoin-gu, Yongin-si, Gyeonggi-do, Korea
Email: woojikchun@gmail.com
Heeyoung Jung
ETRI
218 Gajeong-ro, Yuseong-gu, Daejeon, Korea
Email: hyjung@etri.re.kr
Hong Expires April 19, 2016 [Page 22]
