There was a lot of activity in the planning of new storage technologies this past week with both the Intel Developers Forum and the DISKCON 2007 shows taking place. Plenty of talk about flash taking over the world...which it may someday, but not likely as quickly as initially thought....see Robin Harris on Flash isn't living up to the hype. There was also lots of discussion about SSD (solid state disks), including this excerpt below from Pat Gelsinger, senior Intel vice president and general manager of digital enterprise group, at the Intel Developers Forum in San Francisco last week.
"Today if you go to a big data centre and you're trying to run a big transaction engine, you're asking 'How many hard drives do I need to put in place?' and the decision is not based on capacity but on IOPS; 'How much bandwidth can I feed into the database?' or 'How many transactions can I put through the compute infrastructure?" and "How many hard drives do I need to buy based on the aggregate IOPS to feed that machine?' A lot of those hard drives are only 10 per cent or 20 per cent full and they're there simply for bandwidth not for capacity," he said.
-Source IT Pro
Whether you think that flash will help fill in the gap, or that SSD will finally take off, what is undoubtedly clear is the ringing endorsement and industry recognition that the IT world needs more IOPS.
Rather than debate the pros and cons of various storage media, I think it is more important to look at the need for IOPS and then finding the best way to deliver them.
SSDs are an interesting take on the problem, but ultimately leave administrators with a tool that requires more management instead of less. For example, deploying a SSD requires:
- Identifying which data belongs on the SSD (easier said then done)
- Migrating the data to the SSD
- Creating new backup and recovery procedures for the SSD
- Creating new access policies for the LUNs/volumes on the SSD
- Figuring out what to do when the capacity of the SSD is reached
- and so on....
At Gear6, we see things a little differently. Our caching appliances deliver the IOPS and overall performance needed to manage demanding workloads, particularly those that fit into the category of shared I/O...when many clients need access to the same data set. But by focusing on performance caching for existing storage infrastructure, the management burden decreases significantly.
- Let the appliance automatically determine the most frequently accessed data
- Keep your data in place on your existing storage systems, no manual migration or movement required
- Retain all existing backup and recovery procedures
- Retain existing access and security mechanisms
- Allow one caching appliance to accelerate many terabytes of disk-based data
- Scale cache resources on the fly as needed
The simplicity of this caching approach provides a far smoother transition (in fact, almost no transition is required) for customers in search of performance.