Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation
The improvement of CARMA network model by addition of locality flavor as well as IPv6 support are concerned. The direct experimental evidence of the improvement of the efficiency of the peer-to-peer networks in terms of the network throughput using previously proposed CARMA locality awareness method...
Saved in:
| Published in: | Реєстрація, зберігання і обробка даних |
|---|---|
| Date: | 2011 |
| Main Authors: | , |
| Format: | Article |
| Language: | English |
| Published: |
Інститут проблем реєстрації інформації НАН України
2011
|
| Subjects: | |
| Online Access: | https://nasplib.isofts.kiev.ua/handle/123456789/50540 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Cite this: | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation / G.V. Poryev, V.A. Poryev // Реєстрація, зберігання і обробка даних. — 2011. — Т. 13, № 4. — С. 3-11. — Бібліогр.: 6 назв. — англ. |
Institution
Digital Library of Periodicals of National Academy of Sciences of Ukraine| _version_ | 1859998782984290304 |
|---|---|
| author | Poryev, G.V. Poryev V.A. |
| author_facet | Poryev, G.V. Poryev V.A. |
| citation_txt | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation / G.V. Poryev, V.A. Poryev // Реєстрація, зберігання і обробка даних. — 2011. — Т. 13, № 4. — С. 3-11. — Бібліогр.: 6 назв. — англ. |
| collection | DSpace DC |
| container_title | Реєстрація, зберігання і обробка даних |
| description | The improvement of CARMA network model by addition of locality flavor as well as IPv6 support are concerned. The direct experimental evidence of the improvement of the efficiency of the peer-to-peer networks in terms of the network throughput using previously proposed CARMA locality awareness methods is established. The possible influences of the dynamics of the number of seeding and peering nodes in a swarm (including those directly connected to the test rig) are analyzed and taken into account. Main results indicate average 2,5 % improvement in transfer speed in comparison with clean unbiased transfer modes.
Розглянуто вдосконалену реалізацію моделі CARMA, до якої додано новий клас локальності та обґрунтовано включення підтримки новітнього протоколу IPv6; також отримано пряме експериментальне підтвердження підвищення ефективності однорангових мереж із використанням попередньої версії оцінки локальності за моделлю CARMA. Проаналізовано та враховано вплив динаміки кількості завершених і незавершених вузлів у рої (включаючи ті, що прямо зв’язані з експериментальним стендом). Результати експериментів свідчать про підвищення середньої швидкості передачі на 2,5 % порівняно з немодифікованим режимом передачі.
Рассмотрена усовершенствованная реализация модели CARMA, в которую добавлен новый класс локальности и обосновано включение поддержки новейшего протокола IPv6; также получено прямое экспериментальное подтверждение повышения эффективности одноранговых сетей с использованием предварительной версии оценки локальности по модели CARMA. Проанализировано и учтено влияние динамики количества завершенных и незавершенных узлов в рое (включая те, которые прямо связаны с экспериментальным стендом). Результаты экспериментов свидетельствуют о повышении средней скорости передачи на 2,5 % в сравнении с немодифицированным режимом передачи.
|
| first_indexed | 2025-12-07T16:34:53Z |
| format | Article |
| fulltext |
Математичні методи обробки даних
ISSN 1560-9189 Реєстрація, зберігання і обробка даних, 2011, Т. 13, № 4 3
УДК 004.75; 004.724.2
G. V. Poryev, V. A. Poryev
National Technical University of Ukraine «Kyiv Polytechnic Institute»
Peremohy ave., 37, 02056 Kiev, Ukraine
Improved CARMA Locality Estimation Model
for Peer List Reordering and Its Experimental Validation
The improvement of CARMA network model by addition of locality flavor as
well as IPv6 support are concerned. The direct experimental evidence of the
improvement of the efficiency of the peer-to-peer networks in terms of the
network throughput using previously proposed CARMA locality awareness
methods is established. The possible influences of the dynamics of the num-
ber of seeding and peering nodes in a swarm (including those directly con-
nected to the test rig) are analyzed and taken into account. Main results in-
dicate average 2,5 % improvement in transfer speed in comparison with
clean unbiased transfer modes.
Key words: Internet, distributed networks, peer-to-peer networks, locality,
clustering.
Introduction
Nowadays it is hard to underestimate the influence of information technologies in
general, and the Internet in particular, upon our everyday life and activities. They are
reaching so extreme a level of integration and mobility within our life that it could be
compared to the importance of electric power. Today, both technologies are so crucial
to the progress and the very survival of humankind that virtually no activity is done
without either of them.
Analyzing the modern Internet and its traffic-wise composition one might notice
some drastic change that passed by almost without notice since the beginning of the
21st century. Prior to that, different classes of Internet traffic were distributed more or
less evenly [1, 2] — the majority of traffic consisted of HTTP, FTP, NNTP classes, as
well as some minor classes such as DNS, Telnet, IRC etc.
Exponential growth of Internet users that started in about the same period and cor-
responding growth of the number of Internet-connected devices per average user soon
displayed apparent lack of scalability and extensibility of classical «client-server archi-
tecture» in many scenarios. The situation had resulted in the new internetworking pa-
radigm, now known as «peer-to-peer» networks, or simply P2P. Today P2P is the most
dominant traffic class in every major backbone network according to CISCO VNI (Vir-
© G. V. Poryev, V. A. Poryev
G. V. Poryev, V. A. Poryev
4
tual Networking Index).
Since the emergence of P2P networks, the range, scale and diversity of their use
widened significantly, and the field of application of P2P-oriented systems was notably
extended, too. Being previously considered as a means for file sharing or instant mes-
saging, today’s P2P networks serve as a basic infrastructure for a wide range of innova-
tive application scenarios such as VoIP, multimedia on-demand, software delivery,
massive multiuser environments or online games, and even text messaging, where the
traditional client-server architectures are still dominating.
The original concept behind P2P systems was to relieve load stress on centralized
server farms, mirroring servers and other large data arrays with random access from
multiple external nodes. However, the intrinsic asymmetry of end-user broadband data
links has caused ISPs to increase maintenance and upgrade cost of the «last mile»
hardware in order to keep quality of service steady. Some ISPs had also introduced con-
troversial means to detect and shape forcedly end-user bandwidth for traffic recognized
as P2P, affecting file sharing networks in particular. However, these means are nowa-
days circumvented by protocol obfuscation, encryption and tailoring the connection is-
suance and acceptance queues, as well as using UDP as underlying protocol for data
transport.
For this reason, researchers in the area of P2P systems are aiming to optimize P2P
traffic and consider the inherently clustered nature of the Internet as a potential leverage
mechanism. The general idea is to maximize network throughput inside the particular
network clusters while minimizing the traffic usage between such clusters.
In this paper we have two major goals: a) to improve the previously used CARMA
(see below) model by implementing IPv6 support and additional locality flavor into the
network segment model and b) to test experimentally selective connectivity approach
based on the so-called locality information inferred from CARMA model and determine
whether it has any impact on the typical P2P filesharing scenario.
Previous work
Our previous works [3–5] suggested a locally computed approach for topological
distance estimation that does not rely on third-party non-guaranteed external infrastruc-
tures or unreliable active traffic probing and considered its usage in P2P networks. This
approach was previously named CARMA which stands for Combined Affinity Recon-
naissance Metric Architecture.
In short, we have devised CARMA as three-layered, with the first layer being the
locality awareness expressed in flavors [3], the second layer that uses additional traffic
but does not involve actual P2P communication, and the third layer that requires active
communication to remote parties over a compatible protocol.
As described in [5], CARMA works by initially preloading structural information
from publicly accessible services called Regional Internet Registries (RIRs) and con-
verting it into an internal graphlike data structure. Once loaded, CARMA builds a
model to approximate the Internet topology with some simplifications, resulting in 4
common structural layers (CSLs) as follows: a) IPv4 ranges are divided into b)
subranges but at the same time they also belong to c) Autonomous Systems (ASs),
which are joined into sets called d) AS-SETs or ASSETs.
Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation
ISSN 1560-9189 Реєстрація, зберігання і обробка даних, 2011, Т. 13, № 4 5
From these basic entities, CARMA is then able to define and determine the relative
topological locality of two arbitrary nodes by sequentially find the lowest CSL.
As an estimation result, CARMA produces flavor index ranging from 0 to 6 ac-
cording to the lowest found CSL, namely: «Subrange», «Range», «AS», «ASSET»,
«ASSET-Link», «Backbone» and «Distant». The latter is returned when all tests for the
CSLs have failed to produce one.
Improvement of the structural model
As of 2011, the long-anticipated IPv4 address space exhaustion took place offi-
cially. In February 2011 the IANA allocated the last /8 (formerly Class-A) IPv4 ranges
to the RIRs, whereupon the corresponding local address space exhaustion is anticipated
at the level of LIRs (Local Internet Registries) within several months.
The architectural drawback of IPv4 addressing started to become obvious long be-
fore the address space was nearing its exhaustion, as soon as the number of Internet-
capable devices started to grow exponentially and long after the suboptimal decisions to
reserve Class-A networks for unroutable ranges were implemented.
Since then, numerous measures to amend low address availability were designed
and gained widespread use, among those are NAT, proxy servers and dynamic IP ad-
dress pool.
Meanwhile the efforts to create new IP architecture that would overcome such
problems were started and eventually resulted in the design and standardization of the
IPv6.
Shortly, IPv6 theoretically allows 2128 addresses, i.e. 296 times as many as IPv4. Of
course, as with IPv4, not every possible address is routable, there are reserved ranges
and conventions. Still, the address range capacity is far larger. Even though the recom-
mended practice is to implement IPv6 routing aggregation (i.e. sub-range delegation) as
/64, the remaining 64 bits of address theoretically allow every human being on Earth to
have more than billion devices connected through his private sub-network.
Structurally IPv6 introduces insignificant changes in the protocol itself, removing
unnecessary fields such as header checksum and adding flow labeling and IPSec sup-
port.
The reach of connectivity and points of presence of IPv6 are still underdeveloped,
so that no more than 1 % of Internet users are IPv6-capable today.
Since CARMA model was conceived prior to IPv4 address exhaustion, we initially
did not include any specific support for IPv6 modeling. Now so far as the demand for
IPv6 is expected to grow faster, it is reasonable to investigate what might be done to
include it in the CARMA model.
Our research into the database structure of IPv6 indicates the following:
— аddress range delegation as recorded in delegated database is done primarily
through allocating the /32 segments (i.e. 296 addresses per segment) from RIRs to the
LIRs, although various delegation lengths are seen such as /19, /23, /24, /27, /30 and
others;
— db-inetnum database has its IPv6 equivalent, thereby facilitating sub-range
lookup at the typical scale of /48 network;
G. V. Poryev, V. A. Poryev
6
— db-route database has its IPv6 equivalent, thereby allowing the linkage defini-
tion between IPv6 ASes and allocated ranges and sub-ranges.
The existence and general similarity of the IPv6 data to that of the IPv4 makes pos-
sible to upgrade CARMA network model to support IPv6 with insignificant modifica-
tions.
The apparent lack of dedicated IPv6 specific asset database is explained by the
ASSET linkage being done in IPv4 version for both the IPv6 and the IPv4 ASes.
To sum up, provided all relevant databases are accessible, CARMA implementa-
tion is able to build network structural model for IPv6 with the same level of detail as it
was done in IPv4.
Our recent investigation on the peering practices and the apparent ambiguity be-
tween ASSET membership and actual routing path had led to the necessity to add one
additional flavor to the CARMA flavor set.
Since national carriers sometimes are unwilling to participate either in traffic ex-
change points or establish mutually accountable peering partnerships with other local
ISPs, the actual route path may cross several borders even for the address pair belonging
to the same country and to the model entities formally registered under exchange point
or related ASSET.
For situations like this, we deem it necessary to add one additional flavor of local-
ity between «Backbone» and «Distant» flavors, named «Country», making the total
number of possible flavors equal to 8.
This flavor should be returned by CARMA implementation in cases where IP pair
test yielded negative results for «Backbone» case, i.e. where there are no intermediate
ASSET declared to contain corresponding ASSETs of input IP addresses, but when both
of them belong to the IP ranges declared under the same ISO country code in the dele-
gated database.
Consequently, if ISO codes are still different, the «Distant» flavor is returned.
This flavor will potentially allow leveraging advantages posed by the topological
locality in cases of complicated ASSET linkage and peering conditions but where there
are significant differences between routing path lengths and timings in transoceanic
links in comparison with local or neighboring country links.
Experimental validation
In the scope of this paper we also aim to focus on the core of CARMA technique,
determining the peer query order of neighboring nodes for cluster selection while the
second and third layers are intended to fine-tune the first layer. If the first layer works,
they all will work, too; otherwise it does not matter significantly.
In our previous paper [4] we have conducted a series of experiments aimed to es-
tablish a provable relation between CARMA metric expressed in flavor indices and tra-
ditional distance metric still in use today, namely standard traceroute method. Although
traceroute may show up to 20–30 intermediate hop nodes while the previous version of
CARMA only has 7 flavors of no literal interpretation in terms of physical links or in-
terfaces, the experiments have conclusively indicated that there is indeed a relation, and
by means of mathematical statistics it was estimated as a significant one using ANOVA
method, the value of significance was calculated at 8,6 % and 31,4 % for nationally-
Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation
ISSN 1560-9189 Реєстрація, зберігання і обробка даних, 2011, Т. 13, № 4 7
centered and international swarm samples, respectively. Though these values may seem
low, it should be noted that traceroute method is also greatly affected by the network
switchovers, bandwidth conditions throughout every tracing path etc.
Setup
Experimental results obtained in [3, 4] are indicative of good correlation between
CARMA metric and standard traceroute metric. This is important, but only if we were
to deploy CARMA as a replacement in an already existing solution requiring network
metric. In such scenario it would have been enough to comply with the interfacing
specifications, such as taking a remote address as a parameter and giving flavor index as
a result. Such an approach would still offer an advantage despite the apparently lower
resolution in terms of intermediate hop nodes, because unlike traceroute it not only
lacks the need to use actively measurement traffic but, most important, are capable of
estimating locality between two arbitrary Internet nodes while not running on either of
them. Those results, however, are none the less not to be taken as the proof of the over-
all efficiency of the proposed method, since no direct measurements of either increased
throughput or decreased network latency were obtained in the previous works.
There is a more general goal we now envision for CARMA locality estimation
methods and its possible future derivatives, namely to drive the optimization of the
overlay networks on larger scales, to leverage the inherent nature of P2P networks in
that they are more robust, stress-resistant and reliable if consisting of more nodes. This
somewhat paradoxical property of P2P has always been the cornerstone of the technol-
ogy, allowing it to provide economic benefits where other network architectures de-
mand larger expenses or fail altogether.
It is therefore important to demonstrate experimentally that utilizing CARMA
mechanisms one can actually improve performance of deployed solutions by an appre-
ciable quantity. In this context, we define performance as suggested in [6] by the
throughput relative to the specific communication cost. Since by the nature of P2P and
their overlay networks we are implying that the specific communication cost remains
constant, it is necessary to demonstrate the increase in the average network throughput
in a system employing CARMA mechanisms. It is important, however, to consider the
increase in the network throughput not only as simply floating-window average of
transfer speed (as this is unlikely to increase during bandwidth saturation) but to view it
as а complex parameter depending on, particularly, the apparent acceleration of transfer
in terms of time it takes to reach bandwidth saturation.
Design of the experiment
Prior to the actual measurements, some specific conditions pertinent to the setup
needed to be clarified.
The experiments should be carried out live, on an already deployed, roughly
equally distributed overlay network that uses Internet as the transport infrastructure.
This condition is the consequence to the fact that large scale overlay networks with all
underlying Internet segments and their constantly changing specifics are very hard to be
adequately modeled inside an environment since too many external factors would have
been needed to be taken into account. Those include traffic switchovers, bandwidth
G. V. Poryev, V. A. Poryev
8
variations, noise and good approximation of other traffic classes passing by involved
nodes etc.
Among many P2P implementations we have to select those with performance ex-
pressed as relevant to the throughput. By this constraint, the file-sharing P2P networks
are considered the best fit, since transfer speeds and overhead traffic tolerance are di-
rectly influencing both completion time and network latency. Of all file-sharing P2P
currently in widespread use, we chose BitTorrent as more likely to demonstrate band-
width saturation and reasonable plugin-driven extensibility of its most popular client
software uTorrent.
Considering the two previous constraints, we may not push our experimental setup
to all participating nodes in the BitTorrent overlay. And even if this was somehow pos-
sible, we cannot gain any knowledge of present overlay clustering state, including all
the circuits, links and groupings, except from those peer nodes directly connected to our
experimental rig. Therefore to prove directly the effectiveness of the proposed CARMA
method we aim to demonstrate that in the presence of the sufficiently large overlay, de-
ploying CARMA even on a single node in it improves its local performance in terms
defined above.
It is necessary to exclude throughput patterns observed during initial phase of the
transfer, where not all reported peer nodes from the list are queried and the peer lists are
not yet obtained through DHT and PEX mechanisms, if enabled.
Therefore the design of the experiment is as followes:
— Phase A provides the determination of the repeated pattern of the dynamics of
peer and seed node numbers with respect to the channel saturation given that no other
traffic is running through the experimental node at the time;
— Phase B establishes the observable performance impact of using selective peer
query order according to CARMA metric by the first querying the nodes with the clos-
est distance flavor, compared to the normal throughput dynamics without CARMA.
Experiments and data analysis
The experiments were conducted using a single EM64T-powered computer operat-
ing Microsoft Windows 7, running x64 version of uTorrent 3.1, connected to the Inter-
net through end-user ADSL-link with declared downlink and uplink speeds of 4000 and
1000 KBits/sec, respectively. To exclude the influence of the link congestion and de-
crease the probability of choking (in terms of BitTorrent protocol), uTorrent was set up
to shape bandwidth to 2500 kBits/sec on the average. This provided for the saturation
by the connected peers only. The allocated IPv4 addresses belonged to the Ukrainian
Internet Exchange Point UAIX (AS15645), the IPv6 address was obtained through 6to4
tunnel broker operated by NETASSIST.UA, which generally provided for diversity of
CARMA flavors observed.
The experiments consisted of 20 runs, each run involved reconnecting to the Inter-
net to obtain different IPv4 and IPv6 addresses and clean-up possible noise traffic from
previous experiments; initiating clean download of the same 300 MBytes test file from
BitTorrent swarm consisting of 45 nodes on the average. During the first 10 runs Phase
A tests were conducted to establish the typical behavior of the numbers of total and of
connected peers and seeds (Fig. 1,a).
Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation
ISSN 1560-9189 Реєстрація, зберігання і обробка даних, 2011, Т. 13, № 4 9
a) Node number dynamics b) Transfer speed
Fig. 1. Averaged dynamics of number of seeds and peers in the experimental swarm
These results indicated that although the number of seeds and peers after stabiliza-
tion differed only about 20 %, a new node in the swarm is less likely to connect to
seeder node because of unequal distribution of «interest» depending on the availability
of data, resulting in the seeder node's queue is likely to be overcrowded by regular
peers. Note that in this context the node is considered as a «peer» when it has less than
100 % of published content, regardless of the actual percentage. Since our experimental
node has no parts of the content at the beginning of the download, for it all other nodes
are equal in «interest», therefore the peer list and query order is quickly filled by the
peers. This is observed to happen persistently within the first 3 minutes of download.
However, the presence of the peer in the query order alone does not guarantee immedi-
ate commence of file transfer, and in fact for almost all the duration of the run the num-
ber of actually connected pure «peers» remained low.
Total and connected seeds numbers were observed to match closely within the du-
ration of run, slowly opening download slots for the first 9 minutes before reaching the
preset bandwidth saturation limit, after which the client software decided not to seek
additional download sources, remaining at or slightly below 18 connected seeds.
In general, Phase A experiment indicated that due to the nature of overlay network
functioning the average download run does not reach bandwidth saturation limit in the
first 8 minutes of download, if peer selection process remains random. Fig. 1,b shows
typical transfer speeds with noticeable gap within the first 7 minutes which is consistent
with the behavior of node numbers observed earlier.
For the next 10 runs of Phase B, the behavior of uTorrent was augmented through
BTAPPS interface to the external specialized minifirewall in such a way as to block
connections to the nodes of higher flavor indices until all nodes of lower indics are que-
ried. This was necessary as BTAPPS interface lacked the ability to establish directly or
re-sort the querying order or prohibiting the connection request internally. Fig. 2,b
shows typical transfer speeds with externally augmented peer query order, whereas Fig.
2,a indicates the corresponding node number dynamics.
Apparently, the bandwidth saturation limit is now reached within the first 2 mi-
nutes of download session. We assume that this effect was partially leveraged by larger
number of non-seeder nodes at the very first minutes of download, as shown in Fig. 2,a.
G. V. Poryev, V. A. Poryev
10
At the speed of 2500 KBits/sec the observed difference of about 40 % in transfer speeds
for 6 minutes could lead to potential gain of BL = (2500*0,4*6*60)/8 = 7500 KBytes
which in this demonstrational case was 2,5 % of the total publication size or transfer
time decrease.
a) Node number dynamics b) Transfer speed
Fig. 2. Averaged dynamics of number of seeds and peers in the experimental swarm
Conclusion
Taking into account possible discrepancies between formal ASSET linkage regis-
trations and actual routing path we have modified previously developed CARMA model
by adding additional «Country» flavor to help facilitate prioritization where ASSET
tests fail.
Taking into consideration the IPv4 address space exhaustion we have modified
CARMA model construction algorithms and implementations to include IPv6 entities in
the network model.
We have demonstrated that using the locality metric such as CARMA as a factor in
constructing peer query order in the file-sharing P2P applications could result in per-
formance gain with no additional effort on QoS, hardware modifications or channel res-
ervations along the overlay link path. Although in this case the gain of 2,5 % was rela-
tively low, it may be explained by the singularity of the augmented node among it
swarm neighbors.
Strictly speaking, the purpose of the paper was to prove the viability of the concept
in the real Internet. We believe that the application potential of such technique is not
limited to file-sharing in particular nor in the P2P in general. Any distributed system
built in the framework of overlay network using the Internet as a transport layer could
ultimately benefit from utilizing CARMA to facilitate automatic overlay clustering,
lowering interdomain traffic and decreasing intradomain latency.
1. Frazer K.D. NSFNET: a Partnership for High-Speed Networking: Final Report, 1987–1995. —
Merit Network, 1996.
2. Claffy K. Traffic Characteristics of the T1 NSFNET Backbone / Claffy K., Polyzos G.C., Braun
H.W. — In INFOCOM93, 1993. — P. 885–892.
Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation
ISSN 1560-9189 Реєстрація, зберігання і обробка даних, 2011, Т. 13, № 4 11
3. Poryev G. CARMA-Based MST Approximation for Multicast Provision in P2P Networks /
Poryev G., Schloss H., Oechsle R. — In Proceedings of the Sixth International Conf. on Networking and
Services (ICNS 2010). — IEEE: Cancun (Mexico). — P. 123–128.
4. Poryev G. CARMA: A Distance Estimation Method for Internet Nodes and its Usage in P2P
Networks / G. Poryev, H. Schloss, R. Oechsle // Internation. J. on Advances in Networks and Services. —
IARIA. — 2010. — Vol. 4, N 3,4. — P. 114–128.
5. Poryev G.V. Using the Internet Registries to Construct Structural Model for Locality Estimation
in the Overlay Networks / G.V. Poryev // Реєстрацiя, зберiгання i оброб. даних. — 2011. — Т. 13,
№ 1. — С. 78–86.
6. Алишев Н.И. Развитые методы взаимодействия ресурсов в распределенных системах /
Н.И. Алишев. — К.: Сталь, 2009. — 448 с.
Received 12.12.2011
|
| id | nasplib_isofts_kiev_ua-123456789-50540 |
| institution | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| issn | 1560-9189 |
| language | English |
| last_indexed | 2025-12-07T16:34:53Z |
| publishDate | 2011 |
| publisher | Інститут проблем реєстрації інформації НАН України |
| record_format | dspace |
| spelling | Poryev, G.V. Poryev V.A. 2013-10-23T18:01:24Z 2013-10-23T18:01:24Z 2011 Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation / G.V. Poryev, V.A. Poryev // Реєстрація, зберігання і обробка даних. — 2011. — Т. 13, № 4. — С. 3-11. — Бібліогр.: 6 назв. — англ. 1560-9189 https://nasplib.isofts.kiev.ua/handle/123456789/50540 004.75, 004.724.2 The improvement of CARMA network model by addition of locality flavor as well as IPv6 support are concerned. The direct experimental evidence of the improvement of the efficiency of the peer-to-peer networks in terms of the network throughput using previously proposed CARMA locality awareness methods is established. The possible influences of the dynamics of the number of seeding and peering nodes in a swarm (including those directly connected to the test rig) are analyzed and taken into account. Main results indicate average 2,5 % improvement in transfer speed in comparison with clean unbiased transfer modes. Розглянуто вдосконалену реалізацію моделі CARMA, до якої додано новий клас локальності та обґрунтовано включення підтримки новітнього протоколу IPv6; також отримано пряме експериментальне підтвердження підвищення ефективності однорангових мереж із використанням попередньої версії оцінки локальності за моделлю CARMA. Проаналізовано та враховано вплив динаміки кількості завершених і незавершених вузлів у рої (включаючи ті, що прямо зв’язані з експериментальним стендом). Результати експериментів свідчать про підвищення середньої швидкості передачі на 2,5 % порівняно з немодифікованим режимом передачі. Рассмотрена усовершенствованная реализация модели CARMA, в которую добавлен новый класс локальности и обосновано включение поддержки новейшего протокола IPv6; также получено прямое экспериментальное подтверждение повышения эффективности одноранговых сетей с использованием предварительной версии оценки локальности по модели CARMA. Проанализировано и учтено влияние динамики количества завершенных и незавершенных узлов в рое (включая те, которые прямо связаны с экспериментальным стендом). Результаты экспериментов свидетельствуют о повышении средней скорости передачи на 2,5 % в сравнении с немодифицированным режимом передачи. en Інститут проблем реєстрації інформації НАН України Реєстрація, зберігання і обробка даних Математичні методи обробки даних Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation Удосконалена модель оцінки локальності CARMA для перевпорядкування вузлів та її експериментальна перевірка Усовершенствованная модель оценки локальности CARMA для переупорядочивания узлов и ее экспериментальная проверка Article published earlier |
| spellingShingle | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation Poryev, G.V. Poryev V.A. Математичні методи обробки даних |
| title | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation |
| title_alt | Удосконалена модель оцінки локальності CARMA для перевпорядкування вузлів та її експериментальна перевірка Усовершенствованная модель оценки локальности CARMA для переупорядочивания узлов и ее экспериментальная проверка |
| title_full | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation |
| title_fullStr | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation |
| title_full_unstemmed | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation |
| title_short | Improved CARMA Locality Estimation Model for Peer List Reordering and Its Experimental Validation |
| title_sort | improved carma locality estimation model for peer list reordering and its experimental validation |
| topic | Математичні методи обробки даних |
| topic_facet | Математичні методи обробки даних |
| url | https://nasplib.isofts.kiev.ua/handle/123456789/50540 |
| work_keys_str_mv | AT poryevgv improvedcarmalocalityestimationmodelforpeerlistreorderinganditsexperimentalvalidation AT poryevva improvedcarmalocalityestimationmodelforpeerlistreorderinganditsexperimentalvalidation AT poryevgv udoskonalenamodelʹocínkilokalʹnostícarmadlâperevporâdkuvannâvuzlívtaííeksperimentalʹnaperevírka AT poryevva udoskonalenamodelʹocínkilokalʹnostícarmadlâperevporâdkuvannâvuzlívtaííeksperimentalʹnaperevírka AT poryevgv usoveršenstvovannaâmodelʹocenkilokalʹnosticarmadlâpereuporâdočivaniâuzlovieeéksperimentalʹnaâproverka AT poryevva usoveršenstvovannaâmodelʹocenkilokalʹnosticarmadlâpereuporâdočivaniâuzlovieeéksperimentalʹnaâproverka |