PDA

View Full Version : SD! & HDD copy speed fun


DaleMeyn
03-20-2006, 03:36 PM
This is a bit of curioso concerning SD! cloning/updating speeds and HDD data transfer speeds. One would expect that HDD copy speed would strongly influence All Files copying using Erase then Copy, and that conversely, it would have little influence when using Smart Update on an established clone (because nearly all the time is spent evaluating files rather than copying). The data, however, show both are nearly equally influenced by the HDD copy speed.
I have an iMac G5 iSight, OS 10.4.5, with 2 external HDDs. One is a Micronet Fantom 250GB FireWire400 HDD, the other is a Micronet 250GB FireWire400/USB2.0 HDD. Both have identical Acomdata drive mechanisms and Oxford 911 FireWire bridge chipsets. The Fantom HDD has a 2MB buffer, the Micronet has an 8MB buffer.
First, the HDD copy speeds, copying 417.5 MB of digital photos from the internal Mac HD to each external HDD in turn, just for reference:

Fantom FireWire: elapsed timie = 32.9 sec., copy rate = 12.7 MB/sec
Micronet FireWire: elapsed time = 41.05 sec, copy rate = 10.2 MB/sec
Micronet USB2.0: elapsed time = 75.8 sec, copy rate = 5.5 MB/sec

Notice that the Fantom with a 2MB buffer is actually faster than the otherwise identical Micronet with a 8MB buffer in FireWire mode. Anybody know why that would be the case? Also despite advertised speed figures, USB2.0 real-world copy rates are only a little more than half as fast as FireWire400.

Now the SD! performance, cloning from the internal Mac HD to the Micronet HDD:

1. Copy All Files using Erase then Copy:

Micronet FireWire mode: 13.6 GB cloned in 71m39s, cloning rate = 3.23MB/sec.
HDD copy time: 13.6GB copied @ 10.2MB/s, disk time = 22m13s, 31% of total.

Micronet USB mode: 13.5 GB cloned in 103m33s, cloning rate = 2.22MB/sec.
HDD copy time: 13.5GB copied @ 5.5MB/s, disk time = 40m51s, 39% of total.

2. Copy All Files using Smart Update to an existing clone:

Micronet FireWire mode: 13.5GB cloned in 4m28s, cloning rate = 50.4MB/s.
HDD copy time: 48.0MB copied @ 10.2MB/s, disk time = 4.7sec., 1.7% of total.

Micronet USB mode: 13.5GB cloned in 6m18s, cloning rate = 36.8MB/sec.
HDD copy time: 20.4MB copied @ 5.5MB/s, disk time = 3.7sec, 1.0% of total.

Using the USB2.0 mode slows the overall cloning process by about 45% when doing Erase then Copy, and about 36% for Smart Update. In this trial, Smart Update makes negligible use of actual disk transfer time, but even so, the slowdown using USB instead of FireWire is significant, in a relative sense, nearly the same as when doing a complete Erase then Copy operation, where disk transfer time is a large fraction of the total. It would be more meaningful in terms of drive transfer speed per se if one compared 2 drives, both operating in FireWire400, but with different access rates of the drives themselves. FireWire and USB2.0 are different systems, having very different operation overhead requirements. Thatís probably why FireWire400 is nearly twice as fast for copying as USB 2.0, despite their similar advertised copy rate figures.
I can recommend the Micronet/Fantom HDDs, they are fast, behave very well under OS 10.4.5, and are pretty cheap. How long theyíll last, I donít know, but at present they are performing magnificently, with absolutely no mounting/attaching problems or other mysteries. I had another makerís HDD with the Prolific chipset, it just wouldnít reliably mount or attach in FireWire mode, had to use it in USB2.0 mode to get it to behave reliably. Evidently OS 10.4.5 is designed to work well with the Oxford 911 (FireWire400) and 922 (FireWire800) chipsets. So when selecting an HDD or separate enclosure, make sure it has one of those.

dnanian
03-20-2006, 03:48 PM
Smart Updates does, indeed, spend time evaluating, but evaluation involves a ton of I/O too, Dale -- we're stepping through two drives, and reading data and metadata for every file on the drive. So, it'll definitely be affected by the speed of the I/O interface.

USB is also a unidirectional interface, which has an effect, and can be slowed down by many factors, including the presence of any non-HiSpeed devices on the bus.

And, absolutely, you want an Oxford bridge. The Prolific bridges just don't work reliably. The drives we recommend all have Oxford bridges.

DaleMeyn
03-20-2006, 04:51 PM
It does seem to be the USB2.0 interface that's doing it, alright, and Oxford seems to rule. In light of my results, is there any good reason to prefer an HDD with 8MB buffer over an otherwise identical one with only a 2MB buffer (which would be cheaper, in general)? Dale Meyn.

dnanian
03-20-2006, 05:24 PM
It's hard to say, Dale. Different usage patterns exercise the cache in different ways...