RACS_ A Case for Cloud Storage Diversity - Hussam Abu-Libdeh by bestt571


Cloud storage in the cloud (cloud computing) a new concept of the extension and development of the concept is defined by clustered applications, grid or distributed file system functions in the network a large number of different types of storage devicesapplications together to work together, a common external data storage and business access capabilities of a system. When the core of the cloud computing system operations and handle large amounts of data storage and management, cloud computing systems need to configure a large number of storage devices, cloud computing system is transformed into a cloud storage system, cloud storage is a data storage and management as the core of the cloud computing system.

More Info
									                                                                              RACS: A Case for Cloud Storage Diversity
                                                                                                 Hussam Abu-Libdeh, Lonnie Princehouse, Hakim Weatherspoon

The Cloud Storage Market                                                         Error Correcting Codes                                                         Evaluation

   Cloud storage providers expose simple interfaces to developers. Amazon         RACS uses Reed-Solomon error correcting codes to tolerate failures without     Internet Archives Trace:
   S3’s data model provides flat namespaces (“buckets”) into which named           data loss. Starting with m equal-size disks of original data, we fill k         Trace represents 18 months of activity on the Internet Archive’s FTP sites.
   objects can be uploaded for later retrieval. Other storage services can be     additional disks with redundant data. Any combination of m disks
                                                                                                                                                                                          4.5                                                                                                                                      0.12
   mounted as network filesystems. There is no widely agreed-upon standard         (data or redundant) is sufficient to reconstruct the original data.                                        4
                                                                                                                                                                                                      Inbound Transfers
                                                                                                                                                                                                     Outbound Transfers
                                                                                                                                                                                                                                                                                                                                              Number of Writes
                                                                                                                                                                                                                                                                                                                                              Number of Reads

   interface, but S3’s REST API has been adopted by smaller providers and by      We write (m, n) to indicate that there are n = m + k total disks, any m                                 3.5

                                                                                                                                                                                                                                                                                               Millions of Read/Write Operations
   the open-source Eucalyptus server software.                                    of which are sufficient to reconstruct all original data.                                                  3                                                                                                                                       0.08

                                                                                                                                                                  Data Transfered in TB

   These interfaces differ, but are similar enough to be considered                                                                                                                         2

   interchangeable. Storage providers are forced to compete on price rather                                                                                                               1.5                                                                                                                                      0.04

   than by offering unique services.                                                                                                                                                        1

   Cloud storage is a highly competitive market. These are simplified pricing                                                                                                                0
                                                                                                                                                                                          Aug/2007      Nov/2007          Feb/2008   May/2008      Aug/2008   Nov/2008   Feb/2009   May/2009
                                                                                                                                                                                                                                                                                                                                   Aug/2007      Nov/2007        Feb/2008   May/2008      Aug/2008    Nov/2008   Feb/2009   May/2009

   schemes for the top two cloud storage providers:                                              m=4 Data Disks           k=2 Redundant Disks                                                                                                   Date                                                                                                                                   Date

                                                                                                   (4,6) → "Tolerate up to two failures"
Operation                      Amazon S3                 Rackspace Cloud Files                                                                                   Cost of Hosting on the Cloud: Simulated cost of hosting the Internet
                                                                                  The choice of parameters m and n is a trade-off: Overhead for data storage      Archive’s trace on various cloud storage services.
put / list request             $0.01/1000 requests       $0.01/500 requests∗
                                                                                  and write operations is increased by the ratio n : m. Interestingly, read
get / other request            $0.01/10000 requests      free
                                                                                  operations are not significantly more expensive, since only m disks must be                                                             50
delete request                 free                      free                                                                                                                                                                                    DuraCloud
                                                                                                                                                                                                                                           RACS (m=4,n=5)
                                                                                  read under normal operating conditions.                                                                                                                  RACS (m=6,n=7)
Data transfer in               $0.10/GB                  $0.08/GB                                                                                                                                                                          RACS (m=8,n=9)
                                                                                                                                                                                                                                          Amazon S3 N. CA
Data transfer out              $0.17/GB                  $0.22/GB                 RAID-5 uses a similar strategy to tolerate up to one failure in an array of                                                            40                     Amazon S3
                                                                                                                                                                                                                                            Amazon S3 EU

Storage                        $0.15/GB/month            $0.15/GB/month           hard disks.                                                                                                                                         Rackspace Cloud Files
                                                                                                                                                                                                                                      GoGrid Cloud Storage

  requests are free for files   above 250KB in size                               RACS: Redundant Array of Cloud Storage

                                                                                                                                                                                                            Cost in $K
Why Should We Diversify?                                                          Redundant Array of Cloud Storage (RACS) operates on the same principle                                                                 20

                                                                                  as RAID-5, but rather than using hard disks, it stripes data across cloud
   Cloud storage providers promise high availability, data persistence, and       storage repositories.                                                                                                                  10

   strict impressive SLAs. So why should we diversify storage?
                                                                                                                                                                                                                         Nov/2007           Feb/2008               May/2008              Aug/2008                                                Nov/2008                   Feb/2009                 May/2009
     Outages and Operational Failures:                                                                                                                                                                                                                                                      Date

      Cloud storage providers can experience transient outages. Sometimes
      lasting up to several hours. Diversifying storage improves data                                                                                            Cost of Switching Vendors: Simulated cost of switching cloud storage
      availability.                                                                                                                                              vendors.
      Technical issues at a provider’s site can have unintended consequences.
      In October of 2009 a failure at a Microsoft data center resulted in data
      loss for many T-Mobile smart-phone users.                                                                                                                                                                                      Amazon S3 to Rackspace
                                                                                                                                                                                                                                            RACS (m=4,n=5)
     Economic Failures:                                                                                                                                                                                                                     RACS (m=6,n=7)
                                                                                                                                                                                                                                            RACS (m=8,n=9)

      A change in pricing scheme or the emergence of new competition can                                                                                                                                                 20

      render a particular provider unfavorably expensive compared to its
      alternatives.                                                               The goal of RACS is slightly different than RAID-5. Cloud storage is                                                                    15

                                                                                                                                                                                                            Cost in $K
      Clients may not be able to pick an optimal cloud storage provider           assumed to be much more reliable than hard disks, so data loss prevention
      because the switching cost overrides the desired benefits. Thus, clients     is a much less compelling reason to use error correcting codes.                                                                        10

      experience vendor lock-in if their stored data is large.                    RACS lowers the cost of switching providers, e.g., as a result of economic
      The fundamental problem is that clients have to make an all-or-none         failure.                                                                                                                                5
      decision in switching their data to new providers.                            Only m of all data needs to be moved to leave a vendor.
                                                                                    By reducing the impact of vendor lock-in, RACS increases the leverage of
   Main point: By striping data across multiple providers and adding                customers when negotiating contracts with cloud providers.                                                                           Nov/2007           Feb/2008               May/2008              Aug/2008                                                Nov/2008                   Feb/2009                 May/2009
   appropriate redundancy, clients can tolerate outages and operational           RACS is implemented as an HTTP proxy with the same interface as
   failures, as well as adapt to changes in the economic landscape.               Amazon S3
                                                                                                                                                                                                                                 Created with LTEXbeamerposter http://www-i6.informatik.rwth-aachen.de/~dreuw/latexbeamerposter.php

To top