RACS: A Case for Cloud Storage Diversity Paper by Hussam Abu-LibdehCornell University Lonnie...

RACS: A Case for Cloud Storage Diversity

Paper byHussam Abu-Libdeh Cornell UniversityLonnie Princehouse Cornell UniversityHakim Weatherspoon Cornell University

(1st ACM Symposium on Cloud Computing, SoCC 2010, Indianapolis, Indiana, USA, June 10-11, 2010. ACM 2010)

Presented byChayutra Pailom CS Department, UTSA

What is this topic about?• Motivation: Diversify cloud storage providers • Key issues: The risk on being service on solely particular

cloud storage provider• Solutions: Strip data across vendors • Comparisons: Overhead expense and vendor mobility• Possible improvement: Avoid vendor lock-in, reduce the

cost of switching providers and better tolerate provider outages or failures.

• Number of References: 39 references

Agenda1. Terminology2. Background3. Concepts4. Evaluation5. Conclusion

Related Terminology RAID (Redundant Array of Independent Disks)

Picture: http://www.microcenter.com/random_access/newsletters/06_newsletters/0306/in_the_lab.html

RAID 0 (Striped) Data is spread across all drive volumes. Drive

volumes must be the same size. If one drive fails, all data is lost

RAID 1 (Mirrored) Data is copied to two drives simultaneously.

Mirrored drive volumes must be the same size. If one drive fails, data can still be accessed from

the mirrored drive.

RAID 5 Data is spread across all drives, as is error-correcting

parity information. If one drive fails, data can be reconstructed

from the remaining drives.

RAID 5 (Striping With Parity)

Related Terminology

http://www.vectortech.co.th/technic_002.phphttp://www.dataclinic.com/hard-disk-raid-recovery.htm

Related Terminology RAID (Redundant Array of Independent Disks)

http://www.acnc.com/raidedu/0

Take 1 minute to take this visualization for better understanding

Vendor lock-in A situation in which a cloud customer is “stuck” to his/her

current cloud vendor due to the complexity in switching to another cloud vendor.

you could “unlock” your business from your current cloud vendor – but that not without “sacrifices,” in term of time, money and effort

RACS Redundant Array of Cloud Storage Use RAID 5 as a prototype Disk = Cloud Storage

http://www.cloudbusinessreview.com/2011/05/03/how-to-avoid-cloud-vendor-lock-in.html

Related Terminology

Error Correction Code Use Reed-Solomon error correction code. Oversampling a polynomial constructed from the data. The polynomial is evaluated at several points, and these

values are sent or recorded. Sampling the polynomial more often than is necessary

makes the polynomial over-determined. As long as it receives "many" of the points correctly, the

receiver can recover the original polynomial even in the presence of a "few" bad points.

Used in many different kinds of commercial applications, for example in CDs, DVDs and Blu-ray Discs.

Credit: http://simple.wikipedia.org/wiki/Reed-Solomon_error_correction

Related Terminology

Erasure coding (human understandable version???) The key idea behind a Reed-Solomon code is that the data encoded is

first visualized as a polynomial. The code relies on a theorem from algebra that states that any k distinct

points uniquely determine a polynomial of degree at most k-1. The sender determines a degree k − 1 polynomial, over a finite field, that

represents the k data points. The polynomial is then "encoded" by its evaluation at various points, and these values are what is actually sent.

During transmission, some of these values may become corrupted. Therefore, more than k points are actually sent.

As long as sufficient values are received correctly, the receiver can deduce what the original polynomial was, and decode the original data.

Credit: http://simple.wikipedia.org/wiki/Reed-Solomon_error_correction

Related Terminology

The authors mentioned that cloud storage is more reliable

than hard drives so they leave the readers of this paper assumed less failure of the data and not mentioned that much!

RACS: A Case for Cloud Storage Diversity

Agenda1. Terminology2. Background3. Concepts4. Evaluation5. Conclusion

Background The Cloud Storage Market Cloud storage providers expose simple interfaces to

developers. Amazon S3's data model provides at namespaces (buckets) into which named objects can be uploaded for later retrieval.

Other storage services can be mounted as network file systems. There is no widely agreed-upon standard interface, but S3's REST API has been adopted by smaller providers and by the open-source Eucalyptus server software.

Background The Cloud Storage Market These interfaces differ, but are similar enough to be

considered interchangeable. Storage providers are forced to compete on price rather than by offering unique services.

Cloud storage is a highly competitive market. These are simplified pricing schemes for the top two cloud storage providers; Amazon S3 and Rackspace.

Introduction

Pricing scheme of different cloud storage providers

Background Vendor lock-in Expensive to switch. Subject to possibility of data loss. For business, it is very big issue; go cloud or not. Ways to fix, e.g. Economical approach spread the

data across multiples providers. But high storage and bandwidth cost. Use this paper’s approach!

Background Why should we diversify… Outages and operational failures. Economic failures. By striping data across providers and adding

appropriate redundancy, clients can tolerate outages and operational failures, as well as adapt to changes in the economic matters.

Avoid vendor lock-in!

Background Incentives for Redundant Striping RACS uses Reed-Solomon error correcting codes to tolerate

failures without data loss. Starting with m equal-size disks of original data, we fill k additional disks with redundant data. Any combination of m disks (data or redundant) is sufficient to reconstruct the original data. We write (m, n) to indicate that there are n = m + k total disks, any m of which are sufficient to reconstruct all original data

Credit: http://pubs.0xff.co/posters/racs.pdf

Background Incentives for Redundant Striping (cont.) The choice of parameters m and n is a trade-off:

Overhead for data storage and write operations is increased by the ratio n : m. Interestingly, read operations are not significantly more expensive, since only m disks must be read under normal operating condition

RAID-5 uses a similar strategy to tolerate up to one failure in an array of hard disks.

Agenda1. Terminology2. Background3. Concepts4. Evaluation5. Conclusion

Redundant Array of Cloud Storage

RACS single-proxy architecture

Redundant Array of Cloud Storage

RACS single-proxy architecture

The goal of RACS is slightly different than RAID-5. Cloud storage is assumed to be much more reliable than hard disks, so data loss prevention is a much less compelling reason to use error correcting codes.

RACS lowers the cost of switching providers, e.g., as a result of economic failure. Only 1/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 customers when negotiating contracts with cloud providers.

RACS is implemented as an HTTP proxy with the same interface as Amazon S3.

Redundant Array of Cloud Storage

Redundant Array of Cloud Storage

Multiple RACS proxies coordinate their actions using ZooKeeper

Agenda1. Terminology2. Background3. Concepts4. Evaluation5. Conclusion

Evaluation Internet Archives Trace

Trace represents 18 months of activity on the Internet Archive’s FTP site.

What are the estimated costs associated with storing library type content such as the Library of Congress and the Internet Archive on the cloud?

What is the cost of changing storage providers for large organizations? What is the added cost of a DuraCloud-like (replicate data across multiples cloud providers)

What is the cost of avoiding vendor lock-in using RACS? And what is the added cost overhead of using RACS (depends on the coding configuration = additional data to be written)?

Evaluation

Inbound and outbound data transfers in the Internet Archive trace

for Individual IA data transfers

Read/write requests in the Internet Archive trace for Individual IA

read/write requests

Evaluation Cost of Moving to The Cloud

Estimated monthly and cumulative costs of hosting on the cloud with different storage providers

Evaluation Cost of Switching Vendors

The cost of switching the Internet Archive’s storage provider for various configurations

Agenda1. Terminology2. Background3. Concepts4. Evaluation5. Conclusion

Conclusion The commoditization of cloud services has brought

with it the characteristics of an economy, good and bad. RACS: A simple application of technology to change

the structure of a market. Erasure coding is applied to different types of failure

(economic) than in the storage systems. Microbenchmarks and larger experiments are

simulated for the real-world traces. RACS enables cloud storage customers to explore

trade-offs between overhead and mobility.

Sample References[16] J. Bloemer, M. Kalfane, M. Karpinski, R. Karp, M. Luby,

and D. Zuckerman. An XOR-based erasure-resilient codingscheme. Technical Report TR-95-048, The InternationalComputer Science Institute, Berkeley, CA, 1995.

[30] D. Patterson, G. Gibson, and R. Katz. The case for RAID:Redundant arrays of inexpensive disks. In Proc. of ACMSIGMOD Conf., pages 106–113, May 1988.

[38] M. Vrable, S. Savage, and G. M. Voelker. Cumulus:Filesystem backup to the cloud. Trans. Storage, 5(4):1–28,2009.

Questions?