an data center connectivity

8/18/2019 An Data Center Connectivity

http://slidepdf.com/reader/full/an-data-center-connectivity 1/6

3BIG DATA TRANSPORT

Innovative Data Center Connectivity

BIG DATA TRANSPORT



2BIG DATA TRANSPORT

The last decade has witnessed the rise of social networks, overthe top media distribution, mobile overtaking xed broadbandand many more trends all leading to building of hundreds ofmega data centers around the world. As data centers havecontinued to balloon in size and number, the applications theyrun have outgrown their walls, becoming global content data-bases in the process. The global scale of data center databasesand networks has created the need for truly massive amountsof data to be transported between the various geographicsites. Whether for load balancing, replication, and / or businesscontinuity, the dawn of Big Data Transport has arrived.

Early data center interconnection networks simply made bestuse of whatever network resources and technologies werereadily available. Yet, as the percentage of total transportbandwidth consumed by data centers has begun to eclipse

that of traditional telecommunications, data center operatorshave started to ask for network technology optimized for theirapplication. ADVA Optical Networking understands the uniqueneeds of data center operators and provides products speci-cally designed for each Big Data Transport application.

Efficiency

Just as “beauty is in the eye of the beholder”, similarly net-work “performance” may be different depending upon the ap-plication. For data center operators, efciency is a key part ofnetwork performance. In other words, instead of highest abso-lute performance at any cost, they are seeking highest perfor-mance per Watt, dollar, orsquare meter oor space.Both the size and growthrate of data centers de-mand a close focus on to-tal efciency.

The energy demand ofmega data centers is welldocumented, and the needto locate them near hy-

dro-electric dams and oth-er types of power plantsspeaks volumes. Power isa double hit. Every Watt consumed by equipment must notonly be provided, but thermally dissipated as well. Recently,many data centers have made the decision to intentionally runtheir entire complex at an elevated hot temperature, ratherthan force cooling them to normal ambient temperatures. And,power sometimes brings a triple hit, as additional taxes may

be levied to compensate for the additional burden on local re-sources, infrastructure, and environment.

Space efciency is important as well. Whether leased orowned, campus or shipping container, every square meter ofoor space and rack unit of rack space has a cost associatedwith it. Mega data centers are reaching the limits of what canbe built today, thus shifting the focus from growing bigger togetting more out of what they have already built. Having ex-panded to the limits of what their internal infrastructure canhandle, it is now not uncommon for a data center to run out ofspace before hitting other resource limitations rst.

Scalability

While data centers must plan for exponential and often un-

planned growth, there is the reality of building within today’sscal constraints. Data centers today still provision mostly10Gbit/s and below links. Yet, the percentage of higher speedlinks such as 40G and 100G links is rapidly growing, and 400Gis just around the corner. Which is why future scalability ofany solution they deploy today is so important. All data ratesmay have to coexist on the same ber and transport platformsimultaneously.

FSP 3000 –Efficient, Scalable, Multiprotocol Transport

The FSP 3000 family is the most efcient way of transport-ing data available today. The 300mm depth chassis is half the

depth of most WDM trans-port equipment. Terabits/sof trafc can be transport-ed by a single shelf. Powerlevels below a Watt / Giga-bit/second can be achiev-ed. And our 100G direct-detection technology costsless than running an equiv-alent 10 x 10GbE connec-tion.

The FSP 3000 scales seam-lessly from 10G, 40G,

100G, and soon 400G connectivity. A data center can installa single 10G wavelength today with condence the existingsolution will expand to 192 – 100G channels with no fork liftupgrades or service downtime. And all data rates can peace-fully coexist on the same ber, with no guard bands or awk-ward conguration rules needed.

Data Center Interconnection

Data Center Operator vs Traditional Telecommunications Provider• Telco equipment needs to last decades, while a DCO might replace in fast moving technology cycles (e.g. 3 years).• Telco wants best optical link performance (Bits / s / km / Hz), while DCO wants best performance for given efficiency (Bits /s / W).• Telco’s historically demand 99.999%, “5 Nines”, while a DCO may find 3 nines “good enough”.• Telco’s spec a wide temp range around ambient, while DCO’s spec a narrower temp range and may run hot.• Telco’s rely on rings and mid-span access, while DCO prefer point-to-point and “bookended” amplifiers, at least for now.



3BIG DATA TRANSPORT

Big Data is also big business. The growth of the cloud, datacenters, and content providers has meant Big Data is now asubstantial percentage of the Gross National Product of mostdeveloped nations. But, just as fast as the revenue is createdalmost magically, seemingly out of thin air, it can also disap-

pear overnight unless a business continuity plan is in place.Big Data must not only be stored, but also backed up. In fact,all business data needs backed up, big or small. Whether aSMB or Fortune 500, hospital or utility, a business continuity

plan is key to survival.

Synchronous Mirroring

The total loss of mission-critical data is a highly threateningprospect that can destroy a company’s business future. For

this reason, governments have stepped in with regulations thatseek to protect both businesses, the taxes they generate, aswell as end consumers from loss. Sarbanes-Oxley and HealthInformation Portability and Administration Act (HIPAA) in theUS, and Basel II Capital Accord framework in Europe are buta few examples. Two key components of these regulationsare mandatory geographic dispersal of stored information andminimum data retention periods that can stretch into decades,depending upon the industry. Both require application specictransport networks to satisfy.

The predominate formof data backup todayis synchronous diskmirroring in which aninstant copy of eachnew data set is auto-matically sent to a re-mote site. With thismethod, the storageof a piece of data isnot complete until ahandshake betweenthe main and remote site is complete. Thus, distance and la-tency greatly impact performance. The faster the data pipe,

the lower the latency, the more data can be mirrored in a givenamount of time.

Fast, Low-Latency & Native Multiprotocol Support

ADVA Optical Networking is at the forefront of Storage AreaNetworking (SAN) and Business Continuity technology. TheFSP 3000 has always been rst to support the latest FibreChannel speed steps, the latest being 16GFC with up to 3200Mbytes/sec full duplex throughput. When WAN length distanc-es are involved, the move from 4GFC to 16GFC gives morethan a simple 4X throughput improvement as the higher speedalso reduces time between handshakes. All latency inducingcomponents of the FSP 3000 are optimized for lowest latencypossible so that for a given link distance, throughput is max-imized.

Most data networks today tend to be IP / Ethernet. The pro-tocol translation between Fibre Channel to Ethernet or FCoE

adds additional latency at every point of conversion. The ADVAFSP 3000 is able to natively transport Fibre Channel with nolatency inducing protocol conversions needed. In fact, all SANprotocols are supported including FICON, 4/8/10/16G FC,1/10/100GbE, and SDR/DDR/QDR/FDR/FDR-10 InniBand. Nolonger are rows and rows of equipment needed for translationand transport as the FSP 3000 can natively encapsulate themall into a single efcient, low-latency transport link.

Business Continuity

The ADVA FSP 3000Scalable Optical Transport Solution



4BIG DATA TRANSPORT

The explosive growth in Big Data has been spurred by a rapidmigration to the “cloud”. And for all the hype and buzzwords,when distilled to its basic essence, data has moved from adata owner’s personal hard drive to a hard drive owned bysomeone else. With that move of Big Data to the cloud comesan implicit trust that the data will be just as secure as before.If that trust is somehow broken, then the market has shownitself relentless in response. Fast rising companies have almostdisappeared overnight when breaches in security have come tolight. Securing Big Data has grown just as important as gath-ering it.

Traditionally, cloud providers have relied on encrypting the up-link / downlink between user and cloud. However, recent rev-elations of wide scale eavesdropping have demonstrated thatdata is also vulnerable when traveling laterally between securedata centers. An important piece in securing Big Data, andmaintaining the trust of end users, is to protected it as it trav-els on the physical ber plant.

In-Flight Encryption

In-ight encryption refers to encrypting data at layer 1 in net-work protocol stack. Layer 1 is the physical transport layer ofthe network, where the ber resides. Traditionally, Big Data isencrypted at higher layers in the stack; For example, IPsec isimplemented at layer 3, the packet switching layer. In-ight

encryption is on top of (in addition to) any existing encryptionand may be combined with other methods to increase securi-ty. One unique feature of in-ight encryption is no additionaloverhead is added to data stream. For example Overlay Trans-port Virtualization (OTV), acommonly used IP VPN ser-vice protocol adds 82 bytesoverhead. If a shorter pack-et is transported with OTV,loss of throughput will oc-cur. Most nancial exchang-es have standardized on a

128 byte packet structure.Transporting 128 byte pack-ets over an OTV link will re-duce total link throughputto 60%. With zero over-head, in-ight encryptionmaintains full throughput,completely independent ofpacket size.

Bulk Encryption

ADVA Optical Networking takes in-ight security one step fur-ther by using bulk method encryption. In traditional encryp-tion, the only parts of a packet that are encrypted are relatedheader bits and the payload. Encryption only guarantees thecondentiality of a message, and does not protect against ma-nipulation. It is the job of the checksum to protect the integ-rity of a message by detecting manipulation. In standard en-cryption, the checksum is not part of the encrypted message,allowing the message to be modied and then the checksumupdated. In bulk encryption, the entire packet is encrypted,including the full header and checksum, thus preventing thechecksum from being updated, and guaranteeing the integrityof each packet.

Another huge benet of bulk encryption is being protocol / in-terface agnostic. While other encryption methods only supportEthernet frames, bulk encryption can simultaneously encryptEthernet, Fibre Channel, InniBand, and SONET / SDH into thesame bulk container. A wide variety of data rates and proto-cols can all be aggregated into a single encrypted transportchannel.

100G In-Flight Encryption

The challenge to in-ight encryption is that by operating at thetransport layer, it must also keep up with prevailing transport

data rates. As the Big Data pipes between data centers havegotten fatter, they have also gotten faster. 100G links are be-coming more common. When competing in-ight encryptionsystems topped out at 10Gbit/s, Big Data providers had to

make the difcult choice be-tween running unencryptedat 100G, or link aggregating10 x 10G encrypted ports.Well, Big Data is no longerin that difcult position, asthe ADVA FSP 3000 supportsboth 10G and 100G in-ight,

bulk encryption. Any combi-nation of protocol and datarate can be grouped togetherand transported with 100%throughput over a channel.

Security

Key Benefits of Our Transport Encryption• In-flight encryption secures Big Data where it is most vulnerable• No additional overhead means no loss of throughput with small packet sizes• Bulk encryption includes header and checksum, providing integrity• Any data rate, any protocol can all be aggregated and sent over a single encrypted lane• Encryption manager can be run completely separate from Network Manager



BIG DATA TRANSPORT 5

The amount of data processed in data centers is growing at astaggering 50% CAGR which is driving data centers to evolve.While virtualization and automation enable elastic resourcescaling inside the data center, then the outside data centerconnectivity (WAN) starts to become the bottleneck.

Storage migration and replication, machine cluster migration,and distributed applications all require high speed connectionsto reduce transfer times and increase network throughput. Di-mensioning the WAN for peak capacity leads to an overpro-visioning of the network and an inefcient resource usage.Transport SDN can turn the WAN into a programmable resourceand provide bandwidth-on-demand on a calendared or ad-hocbasis. WAN virtualization allows a resource sharing betweenmultiple tenants. Through integration with cloud orchestration,resources become virtualized both inside and outside the datacenter, working in tandem for truly elastic cloud services.

Controlling Global Networks

As the cloud has continued to grow, two trends have threat-ened to stie that growth. First, the sheer scale of the data-bases and interconnecting networks is staggering, becomingtruly global in nature. The control mechanisms that worked forregional networks do not necessarily work on a global scale.Second, the type of trafc being carried has changed with enduser applications. The majority of trafc now consists of data

ows from caches to consumer, rather than random packetizedinformation between nodes. Both trends are requiring a re-think in the way an IP network should be controlled.

Currently, networks use distributed control in that they maketheir own decisions based upon messages they relay to eachother. Software-De-ned-Networking hasoffered a way of cen-tralizing that controlby allowing routers toreport to data center

hypervisors throughan orchestration layer.It is this orchestrationlayer that promisesto allow the cloud to

continue to grow without bound. But, while centralizing controlis certainly a step in the right direction, if true multilayer op-timization of data ows is to be achieved, the transport layermust be included in this control.

Extending SDN to Transport

The transport layer tends to be complex, analog, and propri-etary in nature. To expect existing SDN controllers designedfor IP switching to handle the extra work load might be askingtoo much. The solution is to abstract the transport network intoa more simplied form that is more easily understood by theSDN controller. The ADVA FSP 3000 Network Hypervisor virtu-alizes the entire transport network and connects to any stan-dard SDN Controller such as an OpenDaylight. The orchestra-tion layer now has full control of data trafc across all networklayers and elements, virtual or real, so that true multilayeroptimization of packet ows can be achieved.

Keeping Up With Growth

The answer to the problem of keeping up with the growth ofthe cloud is not simply building data center networks faster,but rather using existing resources more efciently. Today’s IPnetworks are intentionally provisioned to 30–40% of maximumcapacity to allow for normal trafc variations. Always build-ing and provisioning for peak trafc is a losing proposition. Abetter solution is to make data center networks more exible,

able to respond to both expected and unexpected changes intrafc volume. A recent study by a major data center operatorfound that efciencies as high as 98% could be achieved todaythrough centralized control of end-to-end data ows. Doing soeverywhere would be the equivalent of doubling the size ofthe global cloud and Internet overnight. The ADVA FSP 3000

Network Hypervisor isa major step in thatimportant direction.

Software-Defined Networking

Real World Use Cases for Transport SDN• Proactive bandwidth calendaring lets your network morph to best fit daily trends• Reactive bandwidth calendaring lets you respond to unexpected demand spikes• Follow-the-sun allows global networks to shift resources based on time zones• Automatic virtual machine migration tools can call up new wavelengths as needed to better balance loads across geographically dispersed servers• Get more out of existing network resources, just through more intelligent control



For More Information

ADVA Optical Networking SECampus MartinsriedFraunhoferstrasse 9 a82152 Martinsried / MunichGermany

ADVA Optical Networking North America, Inc.5755 Peachtree Industrial Blvd.Norcross, Georgia 30092USA

ADVA Optical Networking Singapore Pte. Ltd.25 International Business Park#05–106 German Centre

Singapore 609916

[email protected]

About ADVA Optical Networking

At ADVA Optical Networking we’re creating new opportuni-ties for tomorrow’s networks, a new vision for a connectedworld. Our intelligent telecommunications hardware, softwareand services have been deployed by several hundred serviceproviders and thousands of enterprises. Over the past twen-ty years, our innovative connectivity solutions have helped todrive our customers’ networks forward, helped to drive theirbusinesses to new levels of success. We forge close workingrelationships with all our customers. As your trusted partnerwe ensure that we’re always ready to exceed your networkingexpectations. For more information on our products and ourteam, please visit us at: www.advaoptical.com.

ADVA Optical Networking © All rights reserved.

Version 05 / 2014

Optical+Ethernet Innovation • Speed for Customers • Trusted Partner

an data center connectivity

Documents