Cloud Networking Traffic Use Cases and Optimization Opportunities by pptfiles

VIEWS: 0 PAGES: 17

									Cloud Networking Traffic Use Cases
  and Optimization Opportunities
   Greg Bernstein, Grotto Networking
            Fremont, USA
                Motivation
• Understand how network applications make
  significant demands on network resources
• Understand basic constraints and optimization
  opportunities from both the application and
  network stratums
• Understand resilience and reliability
  requirements and how they can be achieved
  across stratum.

                                              2
                      Outline
• End System Aggregation Use Case
  – VoD example
  – Capacity limits and bandwidth trade offs
  – Reliability and Recovery
• Data Center to Data Center (DC-to-DC) Use Case
  – Business scenarios: “cloud bursting”, “rent before you
    buy”
  – Migration and computational arbitrage
  – Reliability and Recovery
• Cross Stratum Interfaces
                                                         3
                           End system aggregation
  • Many clients using services offered at two or
    more “data centers”                                             Client
                                                                      B1       Client                “Region B”
                                                                                 B2     Client
                                                   Data Center 2
                                                                                          B3

                              Client                                                               …           Client
                               A1
                                                                                                                BN
        “Region A”




                         Client
                          A2

                      Client                                                                                       Data Center 3
                       A3                                      Network
                     …
                     Client                                                                                       Client
                      AN                                                                                            C1
                                                                                                      Client
                                                                                                        C2

We use a very                                                                                    Client
                                                                                                               “Region C”
                                       Data Center 1                                               C3
general definition
of a data center                                                             Client     …
                                                                              CN                                                   4
            Common Example: VoD
• Clients
   – Millions of customers for a variety of application providers
• Bandwidth
   – Standard def quality ~ 1.5mbps, HDTV quality ~ 10mbps per
     stream client stream
   – Only 6,666 std or 1,000 hdf streams needed to fill a 10Gbps
     WDM wavelength (assuming no multicast or peer assist)
• Dynamic demand
   – Time of day, day of week, time of year, new releases




                                                                    5
Network Optimization: Optical bypass
 – Here “data centers” are CDN locations
                                                          Client
                                                            B1       Client                “Region B”
                                            Data Center                B2     Client
                                                 2                              B3

                     Client                                                              …       Client
                      A1
                                                                                                  BN
   “Region A”




                   Client
                    A2

                 Client                                                                                Data Center
                  A3                                                                                        3

                …
                Client                                                                                Client
                 AN                                                                                     C1
                                                                                            Client
                                                                                              C2

                                                                                       Client
                              Data Center                                                C3
                                                                                                     “Region C”
                                   1
                                                                   Client     …
                                                                    CN
                                                                                                                     6
        Joint Application/Network
               Optimization
• Application Constraints
  – Streaming capacity of servers in a particular data
    center
• Application Costs
  – Costs of server capacity at different data centers
• Network Constraints
  – Bandwidth capacity between “regions” and data
    centers
• Network Costs
  – Relative cost of bandwidth between a region and
    different data centers
     Above costs and constraints can vary with time!
                                                         7
         Reliability and Recovery
• Server Failure Recovery
   – within a data center  data center internal response
   – between data centers  network response to shift traffic
     to alternative data center
• Network Response
   – MPLS and GMPLS provide the ability to switch “flows”
     quickly minimizing downtime
• Network Failures
   – When network capacity is impacted due to link/node
     failures, network may be able to work with application to
     detour clients to other data centers or “shed load” in a
     controlled fashioned.
• Coordinated Maintenance Operations

                                                                 8
  Data Center to Data Center Networking
                                      Large Data Center




                                      Network




              Business #1                                       Business #N
                  DC                                                DC


                                     Business #2
                                         DC


– For our purposes here we consider a data center any computation facility
  with significant access bandwidth to the network (this does not include
  relatively low bandwidth internet clients)                              9
 Public, Private and Hybrid Clouds


                Sample business cases:
• “cloud bursting”
   – Where a business which owns a data center makes use of
     another data center’s spare capacity to deal with usage
     peaks.
• “rent before you buy”
   – Where a business first deploys a new application in a
     rented cloud before investing in additional data center
     infrastructure. See: “Lessons From FarmVille: How Zynga
     Uses The Cloud”, InformationWeek, May 14, 2011.
                                                               10
Network Access and Traffic Drivers
• Network Access
  – Virtual Private Line/LAN (Ethernet, MPLS, etc..)
    100Mbps-10Gbps +
  – IP transit  General internet access
• Example Application Traffic:
  – Enterprise Resource Planning (ERP)
  – Customer Resource Management (CRM)
  – Financial Transactions
  – Data transfer (Storage backups, VM migration, DB
    replication, etc…)
  – Industry specific applications!!!
 Line Rate Gbps           0.1   1        2.5        10     40      100
 Data transfer per hour   45 GB 450 GB   1.125 TB   4.5 TB 18 TB   45 TB
                                                                           11
       Application and Data Center
       Monitoring and Management




• Detailed application monitoring & management
  – Memory, Disk, Processor, Processes, Database
    transactions, etc…
• Detailed application bandwidth statistics
  – From VM MIBs, sFlow, netFlow, etc…
                                                   12
    Cross Stratum Optimization Opportunities
• Compute Pricing
   – VM instances: Reserved, On Demand, Spot
   – Pricing can vary with demand (spot market) and time of
     day (e.g., due to changing power rates)
• Compute Constraints
   – Limits at local data centers, maintenance windows, …
• Network Constraints
   – Path bandwidth and latency between sites at a given time
• Network Costs
   – Different providers, Congestion based pricing, etc…
                                                  Spot market is where
 Example                                          spare capacity is
 Amazon EC2                                       auctioned off based
 pricing                                          on bids.         13
       Reliability and Recovery
MySQL Cluster for Web
and eComerce
Applications”, Oracle
Corp. White paper.



                                              Independent
                                                                         Independent
                                              Data Center X
                                                                         Data Center Y




                                                                                         Business
                                                                                         #N DC-b
                                                          Network
                        Business
                        #1 DC-a
                                                                                    Business
                                                                                    #N DC-a

                                   Business
                                   #1 DC-b
                                                      Business
                                                                    Business    …
                                                                    #2 DC-b
                                                      #2 DC-a
                                                                                            14
        Reliability and Recovery
• Resilience                                    Line   Data
                                                Rate   Transfer
  – SAN replication                             Gbps   per hour

  – Geographical database replication           0.1    45 GB
                                                1      450 GB
• Data Backup                                   2.5    1.125 TB
  – Recovery Point Objective (RPO – how much 10        4.5 TB
    data may be lost) , Recovery Time Objective 40     18 TB
                                                100    45 TB
• Virtual Machine Migration
  – Data Migration GB to TB to …
  – Machine Size 1-100 GB
                                               EMC


                               Oracle                      15
         Cross Stratum Interfaces
     Applications control/management doesn’t want to deal with network
            details and networks don’t want to always share them!
• Network Query
   – What amount of bandwidth might an application be able to get
     between two or more locations?
   – How much might that cost?
   – Now or some time in the future? Some applications are less
     time sensitive than others…
• Network Reservations
   – On demand or advanced reservations
   – Backup flow reservations
• Recovery
   – Activate pre-reserved recovery flows
   – On demand (perhaps higher cost) recovery action (optimize for
     cost after we have recovered)
                                                                         16
                  Conclusions
• Major Network Traffic Drivers
  – End system aggregation (lots of clients) with data
    centers
  – Data Center to Data Center networking
• Fundamental Tradeoffs for Optimization
  – Fixed and marginal computation and networking costs
  – Computation and networking constraints
• Cross Stratum Interfaces
  – Can apply to both use-cases, make it easier to
    optimize both network and application operations

                                                         17

								
To top