Upload
amoseva
View
223
Download
0
Embed Size (px)
Citation preview
7/28/2019 Bandwidth Myth Wp
1/11
Myths o BandwidthOptimization
Increased use o IT in business, disaster recovery eorts,
cloud computing, data center consolidation projects,and international presence all contribute to corporations
back-end Internet trac growing at an accelerated rate.
But bandwidth growth generally doesnt match the rate o
data transer growthand when it does, bandwidth is still
not a guarantee o throughput improvement.
by Don MacVittie
Technical Marketing Manager
White Paper
7/28/2019 Bandwidth Myth Wp
2/11
Contents
Introduction 3
Common Application Performance Myths 4
Myth #1: Application Perormance Depends Only on Bandwidth 4
Myth #2: TCP Requires Aggressive Back-O to Ensure Fairness 6
Myth #3: Packet Compression Improves Application Perormance 7
Myth #4: Quality o Service Technology Accelerates Applications 8
F5 Brings It All Together 8
Conclusion 11
White PaperMyths o Bandwidth Optimization
7/28/2019 Bandwidth Myth Wp
3/11
White PaperMyths o Bandwidth Optimization
IntroductionMoores Law states that data density doubles approximately every 24 months,
and Metcales Law says that the value o a network grows in proportion to
the square o the number o users. Because these postulates have held true in
practice, global enterprises have ound it advantageous to embed inormation
technology into every aspect o their operations. However, this has led to
increased bandwidth consumptionaccording to Plunkett Research, LTD,
the worldwide data communications services industry now generates revenue
in excess o $3.1 trillion annually.
Despite a growing worldwide thirst or bandwidth, supply has outpaced demand
by a wide margin. During the rapid expansion o the Internet in the 1990s, the
data communications industry created an inrastructure that could deliver cheap
bandwidth in high volumes. In act, bandwidth has become so plentiul that even
the eects o Metcales Law are insucient to consume available capacity or many
years to come. The result o this imbalance has been the commoditization o
bandwidth, rapidly declining bandwidth prices, and a vendor environment that
actively promotes the myth that high bandwidth can address almost any
perormance problem.
But as enterprise application deployments have expanded to the wide area network
(WAN) and increasingly to the cloud, an environment where bandwidth is sometimes
as plentiul as on the LAN, IT managers have noted a dramatic decrease in
application perormance. Many o them wonder, why would two networks with
identical bandwidth capacities, the LAN and the WAN, deliver such dierent
perormance results?
The answer is that application perormance is aected by many actors, associated
with both network and application logic, that must be addressed to achieve
satisactory application perormance. At the network level, application perormance
is limited by high latency (the eect o physical distance and physical communications
medium), jitter, packet loss, and congestion. At the application level, perormance is
urther limited by the natural behavior o application protocols (especially when aced
with latency, jitter, packet loss, and congestion at the network level); application
protocols that engage in excessive handshaking across network links; and serialization
o the applications themselves.
Faster Application andNetwork Access
55% o surveyed customers
selected an F5 solution over the
competition because o aster
application and network access
or users.
Source: TechValidate Survey
o BIG-IP users
TVID: A0A-B7B-546
7/28/2019 Bandwidth Myth Wp
4/11
White PaperMyths o Bandwidth Optimization
Common Application PerormanceMyths
Myth #1: Application Perormance Depends Only onBandwidth
Application perormance and throughput are infuenced by many actors. Latency
and packet loss have a proound eect on application perormance. Littles Law, the
seminal description o queuing theory and an equation that models the eects ophysical distance (latency) and packet loss, illustrates the impact o these two actors
on application perormance.
In an application to networking, this law states:
(throughput) =N (number o outstanding requests)
T (response time)
In terms o IP-based protocols, this translates to:
TCP throughput =congestion window size
round trip time
Thereore, as the round trip time (RTT) o each request increases, the congestion
window must increase or TCP throughput will decrease. Unortunately, TCP does not
eectively manage large windows. As a result, even small amounts o latency and
packet loss can quickly cause network perormance to drop or a given application
to a raction o what would be expected. Even i bandwidth capacity were to be
increased to 100 Mbps, the application would never consume more than 1 percent
o the total capacity. Under these conditions, managers who add network capacity
waste money on a resource that cannot be consumed. The ROI or adding network
capacity that will sit idle is non-existent; more than just capacity is included when
determining perormance.
http://web.mit.edu/sgraves/www/papers/Little%27s%20Law-Published.pdfhttp://web.mit.edu/sgraves/www/papers/Little%27s%20Law-Published.pdf7/28/2019 Bandwidth Myth Wp
5/11
White PaperMyths o Bandwidth Optimization
The Macroscopic Behavior o the TCP Congestion Avoidance Algorithm1 provides a
short and useul ormula or the upper bound on the transer rate:
Rate = (MSS/RTT)*(1 / sqrt{p})
Where:
Rate is the TCP transer rate or throughput
MSSis the maximum segment size (xed or each Internet path, typically 1460 bytes)
RTT is the round trip time (as measured by TCP)
p is the packet loss rate
Figure 1 illustrates this point:
Distance Miles
0 100 1000 10,
450,000
300,000
150,000ThroughputMbps
Figure 1: How TCP perormance is aected by physical distance.
In WANs, sources o high round trip times (e.g., latency) include physical distance,
inecient network routing patterns, and network congestionelements that are
all present in abundance on the WAN.
Today, many TCP protocol stacks are highly inecient when it comes to managing
retransmissions. In act, some stacks may have to retransmit the whole congestion
window i a single packet is lost. They also tend to back o exponentially (i.e.,
reduce congestion windows and increase retransmission timers) in the ace o
network congestiona behavior that is detected by TCP as packet loss. And while
loss is oten insignicant in rame relay networks (less than .01 percent on average),
1 Mahdavi, Jamshid; Mathis, Matthew; Ott, Teunis; and Semke, Jeery. The Macroscopic Behavior o the TCP CongestionAvoidance Algorithm. Computer Communication Review, a publication o ACM SIGCOMM, volume 27, number 3, July1997. ISSN # 0146- 4833. Accessed on November 14, 2011.
http://cseweb.ucsd.edu/classes/wi01/cse222/papers/mathis-tcpmodel-ccr97.pdfhttp://cseweb.ucsd.edu/classes/wi01/cse222/papers/mathis-tcpmodel-ccr97.pdfhttp://cseweb.ucsd.edu/classes/wi01/cse222/papers/mathis-tcpmodel-ccr97.pdfhttp://cseweb.ucsd.edu/classes/wi01/cse222/papers/mathis-tcpmodel-ccr97.pdf7/28/2019 Bandwidth Myth Wp
6/11
White PaperMyths o Bandwidth Optimization
it is very signicant in IP VPN networks that go into and out o certain markets like
China, where loss rates commonly exceed 5 percent, and are oten much higher. In
the latter scenario, high loss rates can have a catastrophic eect on perormance.
When packet loss and latency eects are combined, the perormance drop-o is
even more severe (see Figure 2).
Packet Loss Percentage
40
20
50
30
10
ThroughputMbps
0 2% 4% 6% 8%
HTTP FTP
Figure 2: Sample TCP perormance when packet loss is present.
Myth #2: TCP Requires Aggressive Back-O to EnsureFairness
Many network engineers believe that aggressive back-o in the ace o congestion
is necessary to keep network access air. This is sometimes, but not always, true.
Where congestion control is the responsibility o each host on a networka
network being an environment in which hosts have no knowledge o the other
hosts bandwidth needsaggressive back-o is necessary to ensure airness.
However, i congestion is managed within the network inrastructure by a system
that sees all trac on a given WAN connection, then much greater and moreecient throughput is possibleand aggressive back-o is not required.
Standard protocol behavior species that when hosts consume bandwidth, they
must do so independent o:
The requirements o the application.
The amount o available bandwidth.
The amount o competition that exists or that bandwidth.
7/28/2019 Bandwidth Myth Wp
7/11
White PaperMyths o Bandwidth Optimization
The result is that applications are oten starved or bandwidth resources at the same
time that the network is largely unused. This situation is obviously highly inecient.
A much better solution to the TCP airness problem is to allow individual hosts to
consume as much bandwidth as they need, so long as all other hosts receive
adequate service when they need it. This can be accomplished by implementing a
single congestion window, shared by all hosts, that is managed within the network
itsel. The result is a system in which hosts get the bandwidth they need in periods
o light competition, as well as when competition is more intense.
This single window method delivers consistently higher utilization and greater overallthroughput than aggressive back-o. Hosts each see a clean, ast network that
never loses packets (and thereore doesnt diminish TCP perormancesee Myth #1),
and cumulative trac demands are matched to the overall buering capability o
the network. As a result, IT managers experience optimally utilized networks, under
the broadest range o network latency and loss conditions.
Single window solutions can be constructed so that they are completely transparent to
client systems. Components o such solutions may include TCP technologies such as
selective acknowledgement, local congestion window management, improved
retransmission algorithms, and packet dispersion. These capabilities are then combined
with other technologies that match the throughput requirements o applications to theavailability o network resources, and that track the bandwidth requirements o all
hosts utilizing the network. By aggregating the throughput o multiple, parallel WAN
links, single window technology can achieve even greater throughput and reliability.
Myth #3: Packet Compression Improves ApplicationPerormance
While common packet compression techniques can reduce the amount o trac
on the WAN, they tend to add latency to application transactions and thereore
can impede application perormance. These techniques require that packets bequeued up, compressed, transmitted, decompressed on the receiver, and then
retransmittedall o which can consume substantial resources and add substantial
latency, actually slowing down the very applications that need acceleration.
Next-generation application perormance solutions combine protocol streamlining
with transparent data reduction techniques. Compared to packet-based solutions,
next-generation solutions dramatically reduce the amount o data that needs to be
transmitted, eliminate latency that is introduced by protocol behavior in the ace o
physical distance, and can drive WAN perormance at gigabit speeds. Transparent data
7/28/2019 Bandwidth Myth Wp
8/11
White PaperMyths o Bandwidth Optimization
reduction techniques oten include multiple dictionaries where the level 1 dictionary
is small and highly eective at reducing smaller patterns in data, and the level 2
dictionary is a multi-gigabyte space that can be used to reduce much larger patterns.
Myth #4: Quality o Service Technology AcceleratesApplications
Quality o Service (QoS), i used properly, is a highly benecial set o technologies that
can be helpul or managing application perormance. However, the only thing that QoS
can do is divide existing bandwidth into multiple virtual channels. QoS does nothing to
move more data or streamline protocol behavior. QoS simply decides, in an intelligent
way, which packets to drop. And while it is better to drop packets in a controlled way
than to leave it to chance, dropping packets does not accelerate applications.
Many QoS implementations rely on port numbers to track applications. Because
applications oten negotiate port assignments dynamically, these mechanisms must
be congured to reserve a large port range to ensure coverage o the ports the
application actually uses.
For QoS to be most eective, it should be dynamic. First-generation QoS
implementations reduce large links into multiple smaller links, statically reserving
bandwidth whether it is needed or not. Channelizing a network this way can ensure
bandwidth availability or critical applications like voice, but actually wastes bandwidth
as it is reserved or the specic application, even when the application is not in use.
Dynamic QoS solutions, on the other hand, ensure that bandwidth is reserved only
when applications can use it. One common use o this technology is to extend
enterprise backup windows by enabling continuous data backup when bandwidth
becomes available.
F5 Brings It All Together
F5s WAN optimization solutions deliver dramatic replication perormance and
greatly reduced WAN costs. F5 provides these benets by monitoring the limiting
eects o network conditions, adjusting protocol behavior, and by managing all
levels o the protocol stack, rom the network layer through to the application layer.
Specically, F5 BIG-IP WAN Optimization Manager (WOM) integrates advanced
transport acceleration technologies such as adaptive TCP acceleration, Symmetric
Adaptive Compression, Symmetric Data Deduplication, and session-aware rate
shaping with best-o-class application acceleration technologies including SSL
7/28/2019 Bandwidth Myth Wp
9/11
White PaperMyths o Bandwidth Optimization
encryption ofoading and termination. BIG-IP Local Trac Manager (LTM)and
by extension, BIG-IP WOMis supported by a statistics generation and monitoring
engine that enables organizations to manage application network behavior in real
time. Any communication to a remote BIG-IP WOM device is protected rom prying
eyes with IPsec tunnels, keeping corporate data sae even on the Internet.
BIG-IP Local Traffic Manager+ WAN Optimization Manager
BIG-IP Local Traffic Manager+ WAN Optimization Manager
Internet or WAN
ClientServers
iSession tunnel
Figure 3: Symmetric BIG-IP WOM devices optimize and secure trafc over the Internet.
F5 delivers LAN-like replication perormance over the WAN. BIG-IP WOM optimizes
replication solutions such as Oracle RMAN, Oracle Streams, Oracle Data Guard, and
Oracle GoldenGate or database replication; Microsot Exchange DAG or mailbox
replication; and VMware vMotion over long distances or VM migration, le transer,
and other applicationsall resulting in predictable, ast perormance or all WAN users.
BIG-IP WANOptimization Manager
Data Center 1
01001010011100
11010101001010
10010010010010
01001110110110
Servers
Storage
BIG-IP WANOptimization Manager
Data Center 2
01001010011100
11010101001010
10010010010010
01001110110110
Servers
Storage
Internet orWAN
Figure 4: Accelerate all application trafc on the WAN.
F5 WAN optimization solutions are deployed on F5 hardware, which eatures ault
tolerance, massive scalability, and unparalleled perormance, or on virtual machines.
7/28/2019 Bandwidth Myth Wp
10/11
White PaperMyths o Bandwidth Optimization
For branch oce deployments, BIG-IP Edge Gateway eatures unctionality rom
BIG-IP WOM, BIG-IP WebAccelerator, and BIG-IP Access Policy Manager (APM).
Exchange compression/encryption enable (without BIG-IP WOM)
Deduplication + compression (without Exchange encryption/compression)
Throughput in Mbps (155/ 100/ 0.1%)
0 50 100 150 200
18.15
151.36
Figure 5: BIG-IP WOM improved throughput o Microsot Exchange replication more thaneight times more than Exchange compression and encryption.
NetApp deduplication and compression (without BIG-IP WOM)
BIG-IP WOM (deduplication and adaptive compression)
Duration in Seconds
0 2000 4000 6000 8000
7396
448
Figure 6: NetApp SnapMirror perormance with BIG-IP WOM optimizations shows a 16xperormance boost over NetApp deduplication and compression.
TCP performance with Symmetric Data Deduplication in BIG-IP WOM
When there is a lot o highly repetitive data fowing over the WAN, and that data is
large enough to be caught by the BIG-IP WOM deduplication engine (a conguration
setting controls the denition o large enough), then perormance is massively
improved. This is useul when someone is saving a lot o documents that contain the
corporate logo or other commonly used set o bits.
The key part o deduplication is that while compression works on a single stream,
deduplication can cross all o the streams being transerred across the WAN. This is
important because duplication is oten minimal within stream A, but when all o the
streams being utilized across the WAN are considered, duplication rates are much
higher. The amount o data reduction and the perormance implications o that
data reduction are heavily dependent on a given environment (the amount o
duplication) and the conguration o the BIG-IP WOM device. The larger the
number o bytes that must match, the ewer duplications administrators will receive;
1
7/28/2019 Bandwidth Myth Wp
11/11
but the ewer entries must be checked or a duplication match, and more important,
the longer a given set o bits will be saved or comparison. The smaller the number
o bytes that must match, the aster the cache will ll up and older entries will be
removed, but in theory the more matches administrators will nd.
GoldenGate compression BIG-IP WOM deduplication
Transfer Time in Seconds
0 200 400 600 800 1000 1200
Figure 7: Oracle GoldenGate compression versus BIG-IP WOM deduplication unctions showthat with some datasets, deduplication can massively improve throughput.
ConclusionApplication perormance on the WAN is aected by a large number o actors
in addition to bandwidth. The notion that bandwidth solves all, or even most,
application perormance problems is, simply put, a myth. At the network level,application perormance is limited by high latency, jitter, packet loss, and congestion.
At the application level, perormance is likewise limited by actors such as the natural
behavior o application protocols that were not designed or WAN conditions;
application protocols that engage in excessive handshaking; and the serialization
o the applications themselves.
BIG-IP WOM recognizes the critical interdependence between application-level and
transport-level behavior. It delivers predictable replication perormance and increased
throughput ranging rom 3x to over 60x on networks as diverse as premium-quality,
class-o-service managed networks to commodity, best eortsbased IP VPNs. The
architectural advantages o F5 products result in replication and backup solutions that
deliver best-o-class perormance, massive scalability, and a return on investment that
can be measured in months.
White PaperMyths o Bandwidth Optimization
F5 Networks, Inc.Corporate [email protected]
F5 Networks, Inc. 401 Elliott Avenue West, Seattle, WA 98119 888-882-4447 www.5.com
F5 Networks Ltd.Europe/Middle-East/[email protected]
F5 NetworksJapan [email protected]
2012 F5 Networks Inc All rights reserved F5 F5 Networks the F5 logo and IT agility Your way are trademarks o F5 Networks Inc in the U S and in certain other countries Other F5 trademarks are identif
mailto:info%40f5.com?subject=http://www.f5.com/mailto:apacinfo%40f5.com?subject=mailto:emeainfo%40f5.com?subject=mailto:f5j-info%40f5.com?subject=mailto:f5j-info%40f5.com?subject=mailto:emeainfo%40f5.com?subject=mailto:apacinfo%40f5.com?subject=http://www.f5.com/mailto:info%40f5.com?subject=