RAID arrays

Here is the output from a HP RAID-6 array:

The raw data and configuration for gnuplot is here.

1TB SATA drives

Here is the results of comparing 80G, 160G, and 1T SATA disks. It's interesting to see that the 1T disk can sustain 113MB/s for the outer tracks, it's the first time I've seen a single disk sustain more than about 90MB/s!

One thing to note is that the 1TB disks (from Seagate and WD) are the same size and have just over 10^12 bytes, which is significantly less than 2^40 which is the number that most people think about when the term "tera" is used in the context of computers.

The raw data and configuration for GNUplot is here.

Laptop Hard drives

Here are the results in graphical form comparing the hard drives of an early 1998 Thinkpad 380XD and a late 1999 600E:
As you can see (especially in the case of the 600 series) there are a number of zones giving different levels of performance. The graph for the 600 series shows nice plateaus because I used 30 iterations. From memory I think I only used a small number of iterations for the 380 series, I expect that I would get a nicer graph if I was to redo the test with more iterations.

The number of zones on the disk surprised me. I was only expecting 3 or at most 4.

One thing to note is that the 380xd has a much slower hard drive, but the fastest part of it's drive is almost as fast as the slowest part of the 600e's drive. It is quite easy to imagine scenarios where you can store some data on the past part of a slow hard drive for better performance than the slow part of a fast hard drive!

IDE drives

Below are the results from running ZCAV on a pair of WD3000 disks (300G IDE disks with ATA-100 connectivity). I tested them on a Compaq Evo with 1.5GHz P4 CPU, I only ran one pass because each pass took more than two hours and the results seemed quite consistent (I used a 300M block size). It was running Fedora/rawhide with kernel 2.6.15-1.1996_FC5

The drives can sustain up to 60M/s when running on their own, the red dots on the graph show clear plateaus. The blue and green graphs indicate that there is a system limit on the IO performance of an aggregate of less than 80M/s.

The raw data and configuration for gnuplot is here.

Below are the results from running ZCAV on a pair of 46G IBM drives on an Athlon 800MHz machine with ATA66 running Linux 2.4.x. The red dots indicate the performance of a single disk. The light-blue and pink dots indicate the performance of two disks running at the same time on separate buses (/dev/hda and /dev/hdc). The blue and green dots indicate the performance of two disks running on the same bus (/dev/hda and /dev/hdb). Note that the maximum performance of two drives running on separate buses is about 26MB/s while the maximum performance of two drives on the same bus is about 24MB/s. This indicates to me that there is not much of a bottleneck associated with two disks on the same bus and that the real bottleneck is two disks on the same machine. I'm not sure if this is an issue of OS, motherboard, or just the IDE controller (built in to the motherboard).

CD-ROM and DVD results

In 2000 I bought a Pioneer 105sz DVD drive, it was advertised as a 40* speed CD reader and a 16* speed DVD reader.

For CDs 1* speed means 150K/s over the ENTIRE CD, so 40* would mean 6000K/s over the entire cd! From the following graph you can see that it gets more than 40* speed at the end of the disk, but less than half that at the start. I believe that this is because the drive can't vary the rotational speed much (if at all) unlike the original data CD readers (and probably all audio CD players), this means that the data transfer speed varies across the media. The average speed is about 30* (the next version of zcav will have to calculate this).

Here is a graph of the performance of CD and DVD reading together (I include the CD data for interest because it doesn't obscure the DVD data). You can see that the DVD speed ranges from 8000K/s to 18000K/s. This means that a hypothetical 1* speed DVD would be somewhere between 500K/s and 1125K/s.
HP states that 1* speed DVD is 1353K/s (normally I don't trust information on such spec sheets, but I couldn't find anything better, if someone can inform me of a more authoritative source of such information I would appreciate it). If the HP data sheet is correct then Pioneer's claims of their drive being 16* are slightly exaggerated as the fastest part of the disc is not read at 16* speed!

Note that the slowest part of a DVD is faster than the fastest part of a CD. If you wanted to install a 600M image file onto number of computers then you would be better off using a DVD than a CD for performance reasons.

The raw data and configuration for gnuplot is here.

USB Mass-storage

The graph below shows the results from testing a 250G IDE disk on both the internal IDE interface and on a USB 2.0 interface. As you can see when attached by USB it is significantly slower, even if the USB enclosure could deliver the theoretical maximum performance of about 40MB/s (as it apparently did for 7.5 seconds) it would still be a lot slower than the IDE interface.

The raw data and configuration for gnuplot is here.
Back to Bonnie++ page
Back to ZCAV page
Copyright 2000 Russell Coker