What You Should Know Before Buying a New Hard Drive

The ATA (Advanced Technology Attachment) interface is the most used interface to connect hard drives to PCs, and the ATA interface has been upgraded along the way, but new iterations are compatible with older technology and this makes it susceptible to bottlenecks. The number at the end of an ATA rating corresponds to the interface's maximum throughput. ATA33, for example, has a maximum of 33MBps (Mega Bytes per second), ATA100 a maximum of 100MBps, and so on. This is also true for the new SATA (Serial ATA) interface SATA150 or SATA II SATA300 has a maximum of 150MBps and 300MBps respectively.

You might think that buying a faster drive will increase the speed of the system, well that's to simple. The problem is that the throughput will only be as fast as the slowest link in the chain. If you take an ATA33 or ATA66 drive and hook it up to your new Dual CPU 3.6GHz desktop's ATA133 interface, your system will read and write to that drive, but the process will be limited by the drive's slower rate. Likewise, if you connect an ATA133 drive to a motherboard only capable of ATA100, the drive will only transfer data at a maximum speed of 100MBps.

Below you will find a real world Disk Interface - Drive Size - Transfer Time chart. These results are our own test results from the best equipment we could find in the particular scope of the hardware in question.

 

Interface Type	Throughput	80GB Disk	120GB Disk	250GB Disk
ATA33	670 MBpM	120 Min.	NA	NA
ATA66	1.3GBpM	60 Min.	92 Min.	190 Min.
ATA100	2.0GBpM	40 Min.	60 Min.	125 Min.
ATA133	2.7GBpM	30 Min.	44 Min.	95 Min.
SATA150	2.9GBpM	27 Min.	41 Min.	83 Min.

Ok, so you've bought a new SATA150 drive and you are going to place it in your new Dual CPU 3.6Ghz system now that ought to make a difference right. Well not so fast. While throughput between the controller card and the drive may theoretically run at 150MBps, if the controller card is still only on a 32-bit PCI bus, which has a theoretical maximum of 133MBps. Chip manufacturers like Intel and others do build SATA and SATA II support directly into their chipsets with a 64-Bit bus. These new 64-Bit bus speeds open up the bottleneck. Currently the 64-Bit systems are limited to the SATA II specification or 300MBps.

In the example above upgrading from a 7,200 RPM ATA133 drive to a new 7,200 RPM SATA150 drive you should expect to see about a 3% increase in overall performance. So what are the total limiting factors in achieving maximum throughput in the system you might ask? Well there are a number of factors to consider. CPU speed, BUS bandwidth, Controller burst speed (Burst speeds are not sustainable speeds over the entirety of the disk), Rotational speed of the platter, Cache size, and last but not least file system specification i.e. FAT32, NTFS. Seek speed or overhead is generally nullified by increases in the onboard cache.

CPU speed as a factor in hard drive performance is a principle of its prefetch cache. This either hinders or helps data arriving on the BUS to become synchronized with data leaving the BUS. The bigger the cache and the faster the CPU speed generally speaking, the better (faster) data can come and go to and fro the drive for processing.

The BUS bottleneck we have discussed above, however one thing to check and the most forgotten setting is the Transfer Speed setting. Make sure that the drive is setup to use DMA (Direct Memory Access) as this enables a hardware interface channel between the RAM and Controller. Changing the transfer mode to DMA can increase to performance of the throughput by as much as 60% (Depending on the current setting).

Controller burst speed, is just that burst, the drive can get loaded up as data is read and written and is not a very accurate representation of actual performance.  On our test bench we have seen decreases in the order of 15% throughout the entire sweep of the drive. This number can be subjective, as other factors namely the drives onboard cache affects the loading. As long as all other things are equal i.e. drive speed matches controller interface speed all is well.

Rotational speed of the platter, ATA drives usually run at 5,400, 7,200, or 10,000 RPM while SCSI drives commonly run at 7,200, 10,000, or 15,000 RPM. We will not cover the SCSI interface here, but note that the newer ATA specifications are now very close to that of the SCSI interface and can be said to be statistically identical, at least for our purposes. On our test system we setup two nearly identical 120GB drives, one 5,400 RPM and one 7,200 RPM, we observed an average increase of 6%. Drives that rotate faster perform faster, particularly during disk-intensive tasks, however they are louder and run hotter. Thus, a 10,000 RPM drive requires extra cooling fans and the whole system should be located away from the user. Liquid cooled and liquid bearing drives are a nice solution to the noise problem.

Cache size, is a factor that can increase drive performance. On our test system we tested three commonly used hard drives with cache sizes of, 2MB, 8MB, 16MB, and 32MB. As would be expected, the drive with 32MB cache performed better then any of the others, with an average increase of 5% between each of the 2MB, 8MB, 16MB and an addition 5% increase between the 16MB and 32MB caches.

The File System used can affect the access rate and cache loading. This can be particularly noticeable when using the system with disk-intensive tasks like manipulation of multimedia content. This is a principle of the file system granularity commonly referred to as cluster size. That is the minimum allocation unit on the drive.  Below is a list of common drive sizes, file system types, and the associated default cluster sizes.

Partition Size	Cluster Size (FAT16)	Cluster Size (FAT32)	Cluster Size (NTFS)
1.0 - 2.0 GB	32 KB	4 KB	2 KB
2.0 - 4.0 GB	64 KB	4 KB	4 KB
4.0 - 8.0 GB	NA	4 KB	4 KB
8.0 - 16.0 GB	NA	8 KB	4 KB
16.0 - 32.0 GB	NA	16 KB	4 KB
32 GB - 2 TB	NA	32 KB (*2K/XP)	4 KB

*While Windows 2000 and XP can mount FAT32 volumes larger then 32GB some programs might not be able to see the extra space. In particular to this is the Windows setup program itself which if needed for an in-place upgrade or repair console can certainly be a problem.

Generally speaking, the larger the cluster size the faster the data can be accessed; however when reading smaller files repeatedly the cache can become loaded. There is no one choice when it comes to choosing a cluster size. Unless you are a hardcore media junkie, then the defaults will most likely suit your needs just fine. It is really a matter of trial and error when you veer from the list above. Note that in general that NTFS is the better of the file systems. NTFS will not fragment as fast, it offers average seek time, it will not load down the drive chain caches, and drive size can be anything up to 2TB. If you are tweaking for multimedia and have a dedicated drive then you might try FAT32 with 64KB cluster size.

When all is said and done; if you now have a ATA33 or ATA66 drive and a ATA133 motherboard there is good reason to upgrade to a faster drive, if you are running out of space as well then upgrade now. If on the other hand you currently have a ATA133 7,200 RPM drive and a ATA133 motherboard and space is not an issue you will notice very little improvement when upgrading, unless you use a SATAII-300 controller or are willing to spend the money to get into a 64-Bit MB with Dual CPUs, and the SATAII 300MBps interface (Side note: The 18" cable length limit with the ATA 'Standard 40Pin and 80Pin cables' has been increased to 1m or 3.3' in the SATA and SATA II specification. These round cables also help with airflow in the system and can be routed more easily in bigger tower systems).

Note: We have seen claims that 50 to 60GBpM can be obtained. Beware of these claims.