 |
|
||||
|
Object-based NAS Architectures Eliminate Disk BottlenecksBy Drew RobbJanuary 31, 2005
One solution to this is problem is a more intelligent object-based storage architecture. The idea is to manage data as large virtual objects rather than small blocks. Objects are different than traditional blocks in that they contain both application data and metadata attributes about that data. Thus the resulting storage subsystems become more intelligent by distributing the large portion of the metadata management to the storage subsystem instead of the file system. "An object-based storage system distributes the responsibility for keeping track of the objects in a particular data structure to the actual object itself," said Larry Jones, Vice President of Product Marketing at Panasas, Inc., of Fremont, CA. The object sits directly on the disk. Let's call these object-based disks Object Storage Devices (OSD). An OSD can be created with devices such as disk drives, blades or as part of an array. As it tracks the objects as well as where all of the blocks of data reside, it makes life easier for the application to rapidly access a file. Further, you can scale the system up without compromising performance. How does it work? In traditional storage architectures, as data is stored or retrieved, each block must pass through a single filer head or NFS server to be processed. The control path for metadata is connected to the data path. If too many clients access the data, that single path bottlenecks. An object-based storage system uses a parallel file system to separate the control path from the data path. This can pay big performance dividends when working with large files or datasets. Such technology is more than just a good idea. The Storage Networking Industry Association (SNIA) has had a technical working group for OSD standards for some time. And in September, the ANSI T10 governing body ratified an OSD standard. Further, products have already hit the market that take advantage of objects. Panasas, for example, has developed NAS storage for Linux clusters. Known as he Panasas ActiveScale Storage Cluster, it combines a parallel file system with object-based storage. It delivers scalable bandwidth and random I/O to accelerate application throughput and streamline operations within a single scalable namespace. It is packaged as dense, redundant 5 TB shelves and supports two modes of data access: the DirectFLOW data path and the NFS/CIFS data path. Continued on Page 2.
|
|
|
|
Email
Print
Digg This
Add to del.icio.us
