Data Reliability Tradeoffs

Abdullah Gharaibeh (my colleague at the NetSysLab) has a recent work that explores a combination of heterogeneous storage components in terms of cost, reliability and throughput.

The work proposes a storage architecture that leverages idle storage resources, located on volatile nodes (e.g., desktops), to provide high throughput access at low cost. Moreover, the system is designed to provide durability with a low throughput durable storage component (e.g., tape).

What I like in the solution is that it shows nicely how to decouple components that provide two important features for data-intensive applications: availability and durability. Moreover, this separation (together with the evidence showed by the experiments) helps system administrators to reason about the deployment cost.

Here is the abstract:

Abdullah Gharaibeh and Matei Ripeanu. Exploring Data Reliability Tradeoffs in Replicated Storage Systems. ACM/IEEE International Symposium on High Performance Distributed Computing (HPDC 2009), Munich, Germany, June 2009.

This paper explores the feasibility of a cost-efficient storage architecture that offers the reliability and access performance characteristics of a high-end system. This architecture exploits two opportunities: First, scavenging idle storage from LAN connected desktops not only offers a low-cost storage space, but also high I/O throughput by aggregating the I/O channels of the participating nodes. Second, the two components of data reliability – durability and availability – can be decoupled to control overall system cost. To capitalize on these opportunities, we integrate two types of components: volatile, scavenged storage and dedicated, yet low-bandwidth durable storage. On the one hand, the durable storage forms a low-cost back-end that enables the system to restore the data the volatile nodes may lose. On the other hand, the volatile nodes provide a high-throughput front-end. While integrating these components has the potential to offer a unique combination of high throughput, low cost, and durability, a number of concerns need to be addressed to architect and correctly provision the system. To this end, we develop analytical- and simulation-based tools to evaluate the impact of system characteristics (e.g., bandwidth limitations on the durable and the volatile nodes) and design choices (e.g., replica placement scheme) on data availability and the associated system costs (e.g., maintenance traffic). Further, we implement and evaluate a prototype of the proposed architecture: namely a GridFTP server that aggregates volatile resources. Our evaluation demonstrates an impressive, up to 800MBps transfer throughput for the new GridFTP service

HPDC'08 - Part II

Gilles Fedak delivered an invited talk at the UPGRADE-CN workshop (part of the HPDC'08). He presented the BitDew - a programmable environment that targets data management in Desktop Grids [1].

The rationale behind BitDew is that applications can define routines for data manipulation. These routines are expressed via predefined metadata, which are used by the infrastructure mechanisms to perform data management tasks, such as replication.

In a Technical Report, Gilles and colleagues present use cases and performance evaluation of mechanisms that provide data management functionality in BitDew.

In particular, I found the approach of leveraging metadata interesting. The predefined set of metadata allows the application layer to communicate requirements to the infrastructure regarding the desired level of fault tolerance and transfer protocols, for example.

In fact, this intersects with one of our projects at the NetSys Lab, where we investigate the use of custom metadata as a cross-layer communication method for storage systems [2].

As we use the file system interface to separate between the application and the storage layers, the two approaches (BitDew and our Custom Metadata approach) seem complementary. The metadata passed by the applications via BitDew could propagate to the file system, where it would interact with the Extended Attributes interface (which is exploited by our solution).

More coding fun to come...

[1] Fedak et al. "BitDew: A Programmable Environment for Large-Scale Data Management and Distribution". Technical Report, 6427, INRIA.

[2] Elizeu Santos-Neto, Samer Al-Kiswany, Nazareno Andrade, Sathish Gopalakrishnan and Matei Ripeanu. "Enabling Cross-Layer Optimizations in Storage Systems with Custom Metadata". In HPDC'08 - HotTopics. Boston, MA, USA. September, 2008.

HPDC'2008 - Part I

Last week I participated to two great conferences: the International Symposium in High Performance Distributed Computing (HPDC) and the USENIX. Both events took place in Boston, MA, USA.

There a lot of interesting things to mention. Thus, to avoid a single long post, I will describe a few presentations that I attended (and discuss some ideas) in a series of short posts.

In the first two days at HPDC, there were two interesting workshops: UPGRADE-CN (P2P and Grids for the Development of Content Networks) and MMS (Managed Multicore Systems).

The two workshops had works related to the research projects I am currently working on.

Molina [1] presented his work on designing two protocols that enable collaborative content delivery in mobile transient networks. By transient networks, the authors mean networks composed of devices that are geographically co-located for a short period such as a music festival.

The authors suggest to exploit the multiple network interfaces currently available in most mobile devices and to enable collaborative use of these multihomed devices.

The idea is quite interesting. In particular, it raises some issues from the perspective of distributed resource sharing. It would be good to understand whether incentive mechanisms are necessary in transient networks. The idea is to encourage users to share their connections with a community for collaborative downloading/streaming of content.

On top of that, a nice follow-up work would be to investigate the feasibility of collaborative data dissemination protocols, which are widely used in the Internet (e.g. BitTorrent), in the transient networks scenarios.


[1] Molina et al. "A Social Framework for Content Distribution in Mobile Transient Networks". In UPGRADE-CN'2008.