The PC market today is very full of misinformation. Different companies will have you believe different things, with each company pointing the finger at the others for failings in system performance. If your games run poorly you might have your CPU manufacturer tell you it is the GPU , while the GPU maker says CPU and RAM, the RAM maker cries mainboard and others still throw the accusation at still more components. But where does the problem really lie?
If you take each component as a separate part they all look like great products. We have GPUs capable of processing teraflops of information, CPUs that can handled huge amounts of instructions per second, RAM is getting larger and faster every day, the mainboards we choose are coming with more and more features and the chipsets themselves are faster and more efficient, HDDs are quick and hold massive amounts of data. So just what IS the problem?
Author: Sean Kalinich
Published: August 19th 2008
Type: Editorial (Analysis)
Spelling and Grammatical editor: Planetx64 Staff
Discussion Link
I had a conversation with a storage architect from LSI the other day and it seems that the most common bottleneck in today’s systems is the storage. The common hard drive (and even the fast SAS drives) is simply not fast enough to keep up. (If you do not believe me remove some RAM and let everything page to the HDD; your system will be almost unusable even with a fast RAID setup). I am sure you are thinking that the answer is to move to Solid State Drives. Get away from magnetic storage. Well you would only be partially correct.
Right now SSDs are too fast. Adopting them due to their high I/O will put a huge processing overhead on the majority of onboard SATA controllers. The current models max out the available bandwidth as it is. You are not seeing anywhere near the performance you should be seeing. At that point the controller is the bottleneck.
Just to throw another monkey into the wrench, onboard SATA/RAID controllers use system memory and CPU to control the RAID and SATA functions. The additional information trying to pour through your overworked controller will increase the overhead and can (in limited cases) show a performance decrease unless you configure your system properly (write through over writeback etc)
I am reminded of the transition from ATA33 to 66 to 100 to 133. Each time the standard changed the drive speed saturated the controller and for many of us meant another $150 (or more) for a new controller to handle our new drives.
Today the jump is even more drastic. It is going to require a complete reworking of the way systems handle data, the OSes write and read files, and even the way data is stored. Just about everything will have to be re-tuned to face this new performance improvement. I am talking about everything down to the default block size. But this effort will open up the door to faster and more responsive computing. It will be an interesting ride as the conflict over HDD Vs SSD heats up.