Communications - January 2013

lenges scaling beyond a handful of servers. Still, relational databases have more than 30 years of investment in applications, operations, and skills development that will survive for many years.

The emerging big-data stores as represented by MapReduce and Hadoop offer complementary sets of advantages. Leveraging massive file systems with highly available replicated data, these environments offer hundreds of petabytes of data that may be addressed in a common namespace. Recently, updatable key-value stores have emerged that offer transaction-protected updates. 7 These enormous systems are optimized for sharing with multiple users accessing both computational and storage resources in a prioritized fashion.

Increasingly, these benefits of the big-data environments will be applied to copies of the relational database data used within existing applications. The integration of the line-of-business relational data with the rest of the enterprise’s data will result in a common backplane for data.

The line-of-business department in an enterprise drives the development of new applications. This is the department that needs the application and the solutions it provides to meet a business requirement. The department funds the application and, typically, is not too concerned with how the application will fit into the rest of the enterprise’s computational work.

The IT department, on the other hand, has to deal with and operate the application once it is deployed. It wants the application, database, and servers to be on common ground. It needs to integrate the application into enterprise-wide monitoring and management.

This natural tension is similar to that seen between property developers and the city planning commission. Developers want to construct and sell buildings and are not too fussy about the quality. The city planners have to ensure the developers consider issues such as neighborhood plans and whether the sewage-treatment plant has enough capacity.

As we move to cloud-computing en-
vironments in which the application,
database management system, and
other computing resources are hosted
on a common collection of servers, we
will see an increase in the standards
and expectations over how they will tie
to the enterprise.

Conclusion

The constraints in an environment are what empower the services. The usage pattern allows for supporting infrastructure and concierge services.

Shared buildings become successful by constraining and standardizing their usage. Building designers know how a building will be used, even if they do not know who will be using it. Not everyone can accept the constraints, but for those who do, there are wonderful advantages and services.

The standardization of usage for computational work will empower the migration of work to the shared cloud. With these usage patterns, supporting services can dramatically lower the barriers to developing and deploying applications in the cloud. Lower-level standards are emerging with VMs. These support a broad range of applications with less flexibility for sharing. Higher-level PaaS solutions are nascent but offer many advantages.

We must define and constrain the
usage models for important types of
cloud applications. This will permit en-
hanced sharing of resources with im-
portant supporting services. The new
PaaS offerings will bring tremendous
value to the computing world.

Related articles on queue.acm.org

Commentary: A Trip Without a Roadmap Peter Christy

http://queue.acm.org/detail.cfm?id=1515746

Fighting Physics: A Tough Battle Jonathan M. Smith http://queue.acm.org/detail.cfm?id=1530063

CTO Roundtable: Cloud Computing January 10, 2009 http://queue.acm.org/detail.cfm?id=1551646

References

1. amazon Web services; http://aws.amazon.com/. 2. app Engine; http://code.google.com/appengine/. 3. armbrust, M., Fox, a., Griffith, R., joseph, a. d., Katz, R.

H., Konwinski, a., lee, G., Patterson, d. a., Rabkin, a., stoica, I. and Zaharia, M. above the clouds: a berkeley view of cloud computing; http://www.eecs.berkeley. edu/Pubs/techRpts/2009/EECs-2009-28.pdf.

4. deCandia, G., Hastorun, d., jampani, M., Kakulapati, G., lakshman, a., Pilchin, a., sivasubramanian, s., Vosshall,P. and Vogels, W. dynamo: amazon’s highly available key-value store. ACM Symposium on Operating Systems Principles; http://www. allthingsdistributed.com/files/amazon-dynamo- sosp2007.pdf.

5. Force.com; http://www.force.com/.

6. lakshman, a. and Malik, P. Cassandra—a decentralized structured storage system. Large-scale Distributed Systems and Middleware; http://www. cs.cornell.edu/projects/ladis2009/papers/lakshman- ladis2009.pdf.

7. Peng, d. and dabek, F. large-scale incremental processing using distributed transactions and notifications. In Proceedings of the 9th Usenix Symposium on Operating Systems Design and Implementation (2010); http://research.google.com/ pubs/ pub36726.html.

8. Riak; http://basho.com/products/riak-overview/

Pat helland has worked in distributed systems, transaction processing, databases, and similar areas since 1978. He was the chief architect of tandem Computers’ tMF (transaction Monitoring Facility), which provided distributed transactions for the nonstop system. at Microsoft, he served as chief architect for Microsoft transaction server, sQl service broker, and a number of features within Cosmos, the distributed computation and storage system underlying bing. He recently joined salesforce.com and is leading a number of new efforts working with very large-scale data.

this paper was written before Helland joined salesforce.com and, while there are many similarities, this is not intended to be a description of salesforce’s architecture.