Thank you for the kind words Dave. I try my best.
On the disks ...
For the OS disk, the rotational speed is of the least importance. Access time is and next throughput. So, this is a bit contrary to what Russ says about it and in the end it is up to you (all) to decide yourself;
OS files are small. They are scattered all over, and nothing will be in a nice sequence. So, the disk head needs to go there fast. In the first place this means that the
average access time needs to be high. But what does that mean ?
Acces times of disks have been around 8 ms forever. So, 25 years ago not much different than today. But, today disks are 2TB (etc.) while 25 years ago they were 20MB. So, the
absolute acces time went up with a factor of 100.
Average Acces Time : The time the disk head will reach the "cylinder" it needs
assuming the disk is full. Cylinder : circles on the disk you could see as a track.
Say the data you need is on one cylinder (and the small files most easily are - even a music track will be); it will be on a small portion of it. Once the cylinder is reached, the data blasts through at the rotational speed. Thus, once reached, 10000 RPM will read twice as fast as 5000 RPM. True. Also, 3GB/s will read twice as fast as 1.5GB/s (this is somewhat more complicated, but for argument's sake it is). But :
A disk with the size of 74GB and an average access time of 8ms will take 10 times longer to reach the cylinder than a 740GB disk with 8ms average access time. Why ? because we now sneakily assume the 740GB disk is occupied with the same amount of data as the 74GB disk contains; we just bought the larger disk for the purpose ... And so net, the 740GB disk shows an average access time of 0.8ms over the number of cilinders it ever needs to reach (which is 10 times less than its size, assuming 74GB is occupied).
Aha.
When a disk rotates at 10000 RPM, one rotation takes 1/10000 sec = 0.0001 sec. So, for now assumed the full data on one cylinder has to be read, that takes 0.0001 sec to do it.
Obviously, the 5000 RPM disk takes 0.0002 secs to do that. A difference of 0.0001 sec. Okay ...
1ms = 0.001 sec. So, when the access time of 8ms, which is 0.008 sec is brought back to 0.0008 sec because of what's explained above, this gives a difference of 0.0072 sec.
Gain on acces time = 0.0072 sec
Gain on 2x RPM = 0.0001 sec.
-> The gain on access time is 72 times better.
Yahoo.
Do notice :
This counts for the small files which are scattered over the cyminders occupied. This does not count *at all* for the large (music) files which a. can be read in nice sequence for their data (when we nicely wrote them in sequence) and which b. may be scattered over the cylinders just the same, BUT which is not relevant because we access such a music file only once per 5 minutes. Here the sheer throughput counts, which is a derival of the RPM (again, there's more to it here).
What can further bring down the access time at one moment (which can be seen as the absolute access time) ? the cache the disk carries. Cache : a sort of copy of the data where it belongs originally, but with faster access time; direct accessable memory is an example of it). So, the larger the cache of the disk (which is directly accessable memory), the more chance there is that the data we need is in there. For example, the data of a whole cylinder could go there, never mind we needed a small portion of it only. BUT if, in a next round, a subsequent data portion of the cyinder is needed, it will be in the cache with a. an access time of virtually zero and b. a throughput which does not depend on the rotational speed of the disk.
When we access our music folders to only sum up what's in there (like the Library Area does that), things become more complicated. But because I'm even more smart than you thought
here the Galleries come into play. So, these only contain the small files as meta data + Coverart files, and here the same applies as the OS disk, but more. So now think why I advise to use an SSD for that ? well, because of its unlimited small access time, which btw is absolute here and not "average". This is how my 28000 albums in there show up in less than 10 seconds when "Search" is used (when they are stored as Saved Result Lists this is 0.1 sec or so). Thus, 28000 albums are not only found on "disk" in less than 10 seconds for their more than 560000 files what it comes down to but all the processing is done to cough up the proper (multi volume) structure and the Coverart to show in the Library Area, which latter actually takes the most of the time.
And then to think that I am using an old 1.5GB/s SSD for it, while today they exist in 6GB/s format (real SATAIII).
In the end there is a LOT more to it, like taking care that the one disk is not in the way of the other, when two are used at the same time (like with copying). There is not much in XXHighEnd where *this* applies, but when I'm copying disks I really take care of this (it could be a matter of copying a 2TB disk in 3 hours vs 48 hours, really).
Of course this is all related to the block sizes as well (also see
A Guide to Glitchless Playback), so when you're at it anyway, takes this into account too.
With this little knowledge it may be more easy to see what we need. Of course it is obvious that the largest disk with the best average access time and the highest throughput plus highest rotational speed, will bring us the best all over. But :
It is not as simple as this, because, for example (!), 3TB disks at 1000RPM do not exist. Maybe 1TB disks do, but they will be twice as expensive as the 3TB. This, while it was layed out that the 3TB will gain on net throughput for at least the small files because of the more gain on the access time.
Also, a 64MB cache disk will be faster than an 8MB one. But, you may not be able to find it, and when you do they may be 50% more expensive at least.
While 3GB/s disks exist all over, the 6GB/s do not (yet). So, you may be able to find them in the size you want, but with a poor amount of cache. Or just too small.
Etc. etc.
The "green" disks may not be what you want at all, because they have variable rotation speed and don't ask me at this moment how that will suffer the throughput (for what we do with it !).
A notice which applies to a somewhat different area, but as important : use the proper "disk interface". So, a NAS is totally worthless for its (not) speed, as USB2 is unusable (Firewire the same). eSata is also not "it". Only internal SATA is, and today USB3 is there to our luck. IOW, you can have the disks organised the best as you could think of, but put them in a NAS and you did all for nothing. It looks convenient, but only USB3 really is, net.
Peter
