1, 2, 4, 8, 10: The Evolution of Fibre Channel Page 2
10 Gb Means Big Changes
10 Gb technology, on the other hand, requires a complete infrastructure change and is not backwards-compatible with 8 Gb technologies. I am aware of a few vendors building 10 Gb RAID products that provide two interfaces and two different interface chipsets to cover their bets on which technology will be deployed. This is a smart move, since I believe the winning technology could be a real horse race. There are advantages and disadvantages to each of the choices.
Another point to consider is how long 2 Gb technology will be supported and how long it will be useful for many environments.
Tradeoffs and Choices
For the most part, no one streams data at 200 MB/sec full duplex (2 Gb hardware translated into storage numbers). Some exceptions are: video, imaging and editing and delivery of that data; real-time data capture from satellites for weather and other applications, and real-time data capture for financial services and point of sale.
Bandwidth-type applications are few and far between compared to the standard high I/O per second (IOPS) required by most applications, and many file systems randomize the data so that bandwidth applications are not really running at full rate anyway. The point here is that in many cases, 2 Gb Fibre Channel is not even close to running at the 2 Gb full duplex rates. Add to this the fact that we are getting less bandwidth per gigabyte of disk space, and what does 4 Gb Fibre Channel buy in the real world?
The following chart shows bandwidth per gigabyte of disk space from 1977-2005:
The performance per gigabyte of storage has also been going in the wrong direction. With more data storage and less performance for each device, this can present a significant problem.
Clearly the rate of increase in storage density has not matched the rate of CPU increase, nor has the bandwidth per GB of data, and this change is small compared to bandwidth, since seek and latency times have not changed very much, nor are they going to anytime soon with current technology. Why is this an important consideration for choosing between 1, 2, 4, 8 and 10 Gb Fibre Channel?
New 4 Gb, 8 Gb and 10 Gb technologies will improve the number of IOPS that can be outstanding to most devices. Since RAID controllers often sort I/O requests based on the order of the block addresses of the data, having a larger command queue allows more commands for more devices and improved seek order optimization. This reduces the seek distance and the total time for seeks. Take the following example for 2 Gb Fibre Channel:
Let's say the RAID controller can support a command queue of 512 commands, and that controller is managing 100 disk drives. If the controller is running all disks with RAID-1, then you can look at the 100 disks as 50 disk drives in terms of the number of commands to devices. Given this, you could have on average 10.24 commands per disk outstanding for the controller (512 commands for the controller/50 disk drives). This is not a great deal of commands to sort for seek order optimization given the time to do a seek. On the other hand, if the command queue was much larger, say 2048 commands, as is expected with 4 Gb, you would be sorting 40.96 commands per drive, which could improve the seek ordering, reducing the total number of seeks and improving overall performance. That is not to say that we will now be running IOPS at 4 Gb, but the total channel usage will likely improve, given the reduced time to do seeks.