Understanding MFT
  
Free Space Important
  
Optimizing Drive Life
  
Assuring Drive Life
  
Journals and Logs
  
How fast is fast enough?
  
When is MFT less than perfect?
  
MFT for Laptops
  
MFT for Servers
  
MFT Pricelists
  
Drive Appraisals
  
Downloads
  
Literature
  
Technical Notes
  
Resellers

How Fast is Fast-enough?

Well made Flash drives vary in performance from 2,000 to 14,000 random reads a second. Similarly, the effective write speed of MFT can range (in an 8k block size from 3,700 to 12,500 writes a second. What is the right speed for you? Here, there are two questions that need to get answered. The first is what speed do you need? The second is where is the next bottleneck?

The first question to be answered is the one of bottlenecks. If you have a CPU that is currently at 100% utilization, getting faster flash drives is not going to help at all unless you can also get a faster CPU, or if your work can function in a multi-processor environment and you can get more CPUs or CPU cores. This is not an issue for all people. In the last twenty-five years, CPU speeds have increased 5,000-fold, while disk-drives have only accelerated ten-fold. But it is an issue you will have to look at, especially if you are using ancient equipment.

A related issue applies to Disk Controllers. If you are using single drives directly connected to your motherboard, this is not a concern. But if you are using multiple drives in an array, it is. We have found that the fastest disk controllers are generally the dumbest. These can reach a maximum performance of about 50,000 4kb random reads a second in a highly multi-threaded environment. (Read our article about RAID testing of Mtron drives here.) Conversely, intelligent controllers (i.e. those that have a CPU on the card) only perform at about 60% to 70% of this peak level. The reason is because the controllers spend time thinking about what they want to read before they actually read it. The key point here is that there are practical limits to read and write times based simply upon the practical capacity and throat size of the controllers involved. Accordingly, there is often little advantage, beyond certain thresholds, to building an array of smaller or faster drives.

The same issue applies when engineering storage appliances. Not only are there limits at the controller level, but there are also limits that can apply based upon the latency time of Telnet in iSCSI environments, whether using 1GIGe or 10GIGe connections, as well as fiber channel connections.

The related question is how much speed do you need? If you are using a 15k rpm drive, you are only getting 250 IOs per second out of your current drives. If you are using a 7200 rpm drive, you are not getting much more than 100 IOs per second. Even the slowest drive at 2,000+ IOs is going to give you a dramatic disk-speed improvement. This is likely to be more than enough for most Laptop and Workstation environments, as well as for many servers. Our general caution, then, is to think about how much speed you need to engineer to be happy, and also to be cognizant of the next bottleneck that will occur after you have speeded up your disk sub-system.