Tuesday, September 2, 2008

Pergamum – Storage Clustering takes aim at Disk Based Archival Storage.

Disk based archival storage is increasingly becoming identified as a class of storage that breaks the mold of traditional thinking in storage array design. Disk based solutions have already established a position over tape and optical as the choice for active archive data storage and when the design criteria for cost effective disk based-archive storage is met, disk based solutions will increasingly replace tape and optical as the broader archive storage solution of choice. Considering that 60% to 80% of data in a data center is persistent or archival data and with growing availability demands, the market for such a solution is huge.

Innovative thinking and clustered disk storage technology is the latest to take aim at solving the problem of storing massive amounts of archive data and replacing tape as the primary archive storage medium. The challenges however are not trivial. To be successful any disk based archive solution must be easily scalable in capacity, performance and time, must deliver on long term system and data reliability, be cost effective, be energy efficient, have a high storage density and be easy to manage.

However today’s historically biased, conventional thinking has created disk array designs driven by the need for low latency responses, high IOP’s and optimized to provide access to data at all times. These high level, guiding characteristics were established when most disk based data was transactional and by necessity drove expensive architectures, each subject to frequent refresh cycles. While many vendors are positioning traditional disk array’s as archive solutions but using SATA drives rather than FC drives, employing spin down features for power conservation and reducing cost and functionality by limiting cache size and redundancy features the bottom line is that conventional disk solutions do not have the economics (TCO) needed for the long term storage of archive class data. A disk based archival storage solution that will resonate with the end user must be cost comparable with tape, should have low operational costs (energy, space and management), must have a high storage density (>70TB/sq ft raw) and to avoid expensive upgrades and data migration issues it must be easily scalable in performance, capacity and time (evolvable).

Such an approach was discussed in an article published in the latest issue of :login;, the USENIX journal(1). The article is titled “Pergamum; energy efficient archival storage with disk instead of tape” – sound familiar?

The authors are a group of researches from the University of California, Santa Cruz and interestingly enough a technical director from the Advanced Research Group at NetApp. This paper is reminiscent of the paper published by the early MAID innovators from the University of Colorado (2), which launched the concept of MAID, later developed and productized by COPAN Systems. Are we having an early look at some future storage technology that NetApp intends to develop? Would make sense.

Pergamum, as it is described in the article, is a distributed network of independent, file based, storage appliances that have been labeled “tomes”. Client facing, tomes use a file based protocol but a block based one for inward facing, inter-tome communication which is via inexpensive Ethernet switches. Each tome is a self contained sealed unit made up of a SATA drive, a low cost processor which is the local control and command center managing all intra-tome housekeeping, data management, data integrity and manages the dialog within what I can only describe as a “Borg” like topology with other tome resident processors. NVRAM holds the metadata that enables the tomes collaborative existence and an Ethernet controller provides the connectivity. Despite each tome having its own processor, tome power consumption is low and within the capability of power over Ethernet. Each tome consumes 1 watt in the spin down state and 13 watts when running and compared to MAID’s power appetite of 8.14 watts/TB an equivalently sized Pergamum solution has a power tax in the order on 5 watts/TB, including switches. When calculating system power consumption MAID and Pergamum have the advantage in that all inactive drives stop spinning and limits are enforced on the number of drives that can spin concurrently. This is the feature that effectively corals glutinous power demands. To put some perspective on these numbers a 300GB 15k RPM, Cheetah drive consumes 12W, equivalent to 40W/TB drop the rotational speed to 10k RPM and the power draw drops to 10W, approximately 33W/TB and this does not consider other component requirements. Admittedly Pergamum is barely past the conceptual stage and I have made some perhaps bold assumptions but at least the comparison serves to underline the growing awareness of the need to contain the power appetite of a disk based archive storage solutions.

The potential scalability of this approach is impressive as suggested in the configuration described in the article. Consider a 48–port switch supporting 46 tomes leaving two ports free for inter-switch communication. Increasing the switch count enables the capacity to be increased in 1 to 46 tome increments. A 50 switch configuration could support over 2200 tomes (2200 TB). Using the same assumptions as above this configuration would have a power tax of just over $7/TB. Add 10GE inter-switch links and replicate this configuration and you can see the potential.

It is apparent that there is still a lot to be done and some invention to occur to move this innovative work from concept to practical application but it is at least focused on the right issues and if NetApp become convinced of its viability they are the guy’s to make it happen.
This approach is somewhat similar to the FAB (Federated Array of disks) proposed by the folks from HP Labs back in 2003 and what appears to be the underlying technology for their recently announced HP StorageWorks 9100 Extreme Data System (ExDS) a massively scalable NAS (read archive) product.

So it looks like we have two companies who have declared realistic disk based archive solutions, COPAN with their MAID technology and HP with their soon to be available, ExDS/Polyserve.

Until I learn otherwise, and whether Pergamum turns out to be commercially viable or not, I will give NetApp the credit for realizing that traditional thinking not going to produce the revolutionary design needed for a revolutionary disk based archival storage.

(1)“Pergamun; energy-efficient archival storage with disk instead of tape”; Storer, Greenan, Miller, Voruganti; ;login:, Vol 33, Number 3, June 2008
(2) The Case for Massive Array of Idle Disks (MAID); Colarelli, Grunwald and Neufeld; Dept of Computer Science, University of Colorado, Boulder; January 2002.

No comments: