It's not enough to assume that buying a bunch of RAID is going to solve storage issues.
Administrators need to understand the implications of different storage technologies and methods of data access.
There are two fundamental methods of accessing data: by block number, or by filename/byte offset.
Block protocols such as SCSI-FCP or iSCSI are an obvious solution for predictable, streamed I/O for media servers or data backup.
File access protocols such as NFS or CIFS can be more efficient for small, random I/Os -- such as those required by many database, OLTP and corporate file servers.
Block-level access can be faster for small, simple file systems. As the size and complexity of a file system grows, response times get slower as the application server spends more of its time housekeeping and less time actually running the application. The cost -- in terms of CPU and buffer RAM -- required to access a small chunk of application data grows exponentially.
By offloading file I/O to another server, the application server can scale much better. From the application server's perspective, the cost in CPU and buffer space required to access that 100 bytes of application data via CIFS or NFS is always the same -- regardless of the complexity of the target file system.
New technologies will solve different issues.
iSCSI will allow media and backups to stream efficiently across IP networks, but it is not a solution for small, random application or user I/O. Because it is block-based, it will not allow read/write data sharing and it can require movement of large numbers of virtual disk blocks across the network just to get at small amounts of data. It has the same TCP/IP overheads as CIFS and NFS.
DAFS will improve large, predicted I/O while maintaining file-based access and true sharing, but it will not stream as efficiently as iSCSI and still has poor vendor support.
Existing NFS and CIFS file-sharing protocols are currently the most efficient at fine-grained application and user data access, and as more components of TCP/IP are offloaded to hardware they will continue to get faster, but they are still slow for media serving and backup.
NAS devices scale and manage file systems better than application servers, usually due to a dedicated OS and/or superior file system and disk data layout techniques. If you're planning to use NAS, you need to ensure that security and management features are compatible with your application servers.
By looking carefully at their mix of application and user data access requirements, administrators can pick the best mix of technologies to suit their needs - and ease their growth pains.
About the author: This tip was written by Alan McLachlan, a systems engineer specialising in storage management solutions for the Australian systems
integrator, ASI Solutions (http://www.asi.com.au/).
