Discussion of HDD selection philosophies and strategies

[Rajahal]

Anyway, you'll probably be interested of thinking about 'sweet spot' numbers in this scenario.

I am keenly interested in that scenario, as I am looking to build a small 3-5 disk system for a friend.  Let's see if I understand your 'non-algorithm'  

 

(total advertised size of disks in GB * 0.928)/(cost of server hardware not including disks + cost of unRAID license + cost of data disks + cost of parity disk + cost of cache disk, if used) = cost per drive slot in GB/$

 

To achieve 6 TB storage capacity with the cheapest available WD 2 TB disks:

 

(6000 GB * 0.928)/($800 + $69 + (3 * $179.99) + $179.99) = 3.5041 GB/$

 

So about 3.5 GB per dollar spent.  

 

To achieve 6 TB storage capacity with the cheapest available WD 1.5 TB disks:

 

(6000 GB * 0.928)/($800 + $69 + (4 * $109.99) + $109.99) = 3.9240 GB/$

 

So about 3.9 GB per dollar spent.

 

In this scenario, it seems that the 1.5 TB drives end up being the slightly 'cheaper' option (more bang-for-your-buck), though of course the 2 TB drive option would leave one bay unused for future expansion (up to 8 TB).

 

Either way, the NS-520 is one expensive beast!  I estimated the NS-520's cost to be $800 based on this.  However, I cannot find a reliable source for that product anywhere!  Where did you get yours, Purko?  And how much did it cost?  

 

I really wanted one of these, but its looking to not be the best deal, especially compared to LimeTech's new RB-1200.  I guess smaller always has a premium.

 

P.S.:  BTW, I'm glad you kept your composure.  Grin  

Its one of the few things I'm good at  Proofread before posting, I always say (to myself).

[purko]

I really wanted one of these

If you do, just PM me and can find you one for half the price you mentioned.

 

[Rajahal]

Get ready for a monster of a post...it's amazing I ever get anything done at work

 

I really wanted one of these

If you do, just PM me and can find you one for half the price you mentioned.

 I very well may do that, thanks for the offer!

 

I was interested in applying your method to my home-built server.  The primary problem is that I initially built my server with spare parts, and have slowly upgraded it over time.  Luckily, I ordered all of my server's parts from Newegg except for the mobo (which was from Fry's), but I do remember how much I paid for the mobo ($100 for a mobo and CPU combo).  The PSU was also part of a bundle ($80 for an Antec case and the Earthwatts 380 W PSU), so I estimated paying about $40 for it.  The truly wonky bit are the hard drives.  When I first built my server, I used four 500 GB drives, a 320 GB drive, and a 250 GB drive, all of which I had owned for several years.  I'm sure I paid an arm and a leg for them when I originally bought them (in fact, I remember paying close to $300 for the 320 GB drive back in the day!).  Therefore, I simply called each of my three remaining 'original' hard drives as being worth $100, whereas I probably paid a lot more for them at the time.  Anyway, here's the partially estimated breakdown of my server:

 

Gotta love the deals I found on the TX4 and the RAM!

 

Now to plug these numbers into the formula:

(total advertised size of disks in GB * 0.928)/(cost of server hardware not including disks + cost of unRAID license + cost of data disks + cost of parity disk + cost of cache disk' date=' if used) = cost per drive slot in GB/$[/quote']

 

(5500 * 0.928)/(258.94 + 75 + 579.96 + 99.99 + 100) = 4.5821 GB/$

 

About 4.6 GB per dollar spent.

 

Not much of an advantage over the Norco system, actually.  Now let's see what it would look like I built a similar system based on today's prices for the same basic components, but the storage capacity consolidated onto fewer, larger hard drives:

 

Using only 2 TB HDDs:

 

(6000 GB * 0.928)/(276.94 + 69 + 739.45) = 5.1299 GB/$

 

About 5.13 GB per dollar spent.

 

Using only 1.5 TB HDDs:

 

(6000 GB * 0.928)/(276.94 + 69 + 599.44) = 5.8896 GB/$

 

About 5.89 GB per dollar spent.

 

Again, the 1.5 TB HDD-based server appears to win out.  I tried to think of all contingencies (such as adding in the 5.25" to 3.5" drive bracket that may be necessary to mount a 6th drive in that case), but of course I may have overlooked something.  If anyone wants to actually use these Newegg wishlists, just PM me with an email address.

 

Notes: On both of these systems, I used a mixture of brand names to minimize the chance of a multiple drive failure.  This meant using a Hitachi drive on the 2 TB system, which I would normally avoid.  I guess Hitachi doesn't make 1.5 TB drives, or maybe Newegg just isn't stocking them at the moment.  This meant that I had to double up on 1.5 TB WD Green drives in the 1.5 TB HDD-based server.  Again, this is something I would normally avoid.  Also, the motherboard is a bit overkill, but its the closest to my current board that I could find.  The Antec p180, a full tower case, is no longer offered by Newegg, so I subbed in the Antec mini p180, which is one of the finest cases I've had the pleasure of using.  The full tower p180 would be a bit ridiculous for a 4-6 drive server anyway.  I chose the 120 GB WD Blue as a cache drive for both because it is the cheapest WD drive available.  Even in the age of full BluRay ISOs, 120 GBs should be plenty big enough for a general purpose cache drive.

[purko]

(total advertised size of disks in GB * 0.928)/(cost of server hardware not including disks + cost of unRAID license + cost of data disks + cost of parity disk + cost of cache disk, if used) = cost per drive slot in GB/$

 

I never said the above 'quote'.  I have no idea from where you get that 0.928 multiplier, but I am not going to discuss it, since it cancels itself in a comparison. So I'll just go with it.

 

To achieve 6 TB storage capacity with the cheapest available WD 2 TB disks:

 

(6000 GB * 0.928)/($800 + $69 + (3 * $179.99) + $179.99) = 3.5041 GB/$

 

So about 3.5 GB per dollar spent.  

 

To achieve 6 TB storage capacity with the cheapest available WD 1.5 TB disks:

 

(6000 GB * 0.928)/($800 + $69 + (4 * $109.99) + $109.99) = 3.9240 GB/$

 

So about 3.9 GB per dollar spent.

 

In this scenario, it seems that the 1.5 TB drives end up being the slightly 'cheaper'

 

Two things about that:

 

1) I don't know why you chose to use $179.99 again, when just yesterday we checked that 2TB Wester Digitals can be had for $169 shipped.

Moreover, I wouldn't even use $169, because if you just follow the 'good deals' board, you can always buy 2TB disks for $139.

 

2) Again, you chose to leave electricity cost out of the equasion, and yet compared scenarios with different number of disks.

If you insist on leaving electricity out, then (for the purposes of the comparison) you should calculate it with equal number of disks.

 

So, when taking 1 and 2 above into consideration, here is the slightly revised arithmetic:

 

a) 5 x 1.5TB disks:

(6000 GB * 0.928)/($800 + $69 + (5 * $109.99)) = 3.9240 GB/$

 

b) 5 x 2TB disks:

(8000 GB * 0.928)/($800 + $69 + (5 * $139.99)) = 4.7318 GB/$

 

So now it turns out that that with the 2TB disks you'll be getting 20.6% more GigaBytes for your dollar.

( 3.9240 * 120.6% = 4.7318 )

 

Of course, the above calculation is not quite precise either, because that particular box can be obtained for much less than what you assumed.

But that doesn't matter for what I am trying to say here:

The main point is that when you talk about 'sweet spots' you have to take into consideration a lot more things than just dividing disk gigabyte by disk dollar.

 

Purko

 

Edit: clarity.

[Rajahal]

I just noticed one mistake on my previous post.  I included the shipping charges on the RAM, but not on anything else.  Oh well.

 

Since I'm so painfully bored at work today, I decided to keep going with this and evaluate my 'dream' build:

 

Starting out with just 3 TB capacity and 3 empty hot swap bays:

 

(3000 GB * 0.928)/(353.89 + 69 + 369.46) = 3.5135 GB/$

 

About 3.5 GB per dollar spent.

 

Doubling the 1.5 TB drives (2 of each type) for a capacity of 6 TB, all hot swap bays filled:

 

(6000 GB * 0.928)/(353.89 + 69 + (2*369.46)) = 4.7925 GB/$

 

About 4.8 GB per dollar spent.

 

That set up could even support 10 total drives if the included internal 4-in-3 cage were used, and a 4 port SATA expansion card were added.

[Rajahal]

(total advertised size of disks in GB * 0.928)/(cost of server hardware not including disks + cost of unRAID license + cost of data disks + cost of parity disk + cost of cache disk, if used) = cost per drive slot in GB/$

 

I never said the above 'quote'.  I have no idea from where you get that 0.928 multiplier, but I am not going to discuss it, since it cancels itself in a comparison. So I'll just go with it.

Purko, thanks for the responses.  I know you never said that 'quote' directly, that was my interpretation/translation of this:

 

You can figure out the total usable size from data disks,

then divide that by the cost of the server plus cost of all disks including parity plus cost of the license.

Only this way you can have a somewhat realistic estimate for gb/dollar.

 

The 0.928 factor is simply to convert the manufacturer's advertised capacity to the actual usable capacity.  At the GB level, the difference is 7.2%.  I find it easier to multiply by the percentage left over, not what is missing.  100%-7.2% = 92.8%, so 1000 (advertised) GB * 0.928 = 928 (actual) GB.  Perhaps there's a mistake here, though, because the actual number is 931 GB, isn't it?  Oh well, its still pretty close.  Like you said, it cancels itself out when comparing drives anyway.

 

To achieve 6 TB storage capacity with the cheapest available WD 2 TB disks:

 

(6000 GB * 0.928)/($800 + $69 + (3 * $179.99) + $179.99) = 3.5041 GB/$

 

So about 3.5 GB per dollar spent. 

 

To achieve 6 TB storage capacity with the cheapest available WD 1.5 TB disks:

 

(6000 GB * 0.928)/($800 + $69 + (4 * $109.99) + $109.99) = 3.9240 GB/$

 

So about 3.9 GB per dollar spent.

 

In this scenario, it seems that the 1.5 TB drives end up being the slightly 'cheaper'

 

Two things about that:

 

1) I don't know why you chose to use $179.99 again, when just yesterday we checked that 2TB Wester Digitals can be had for $169 shipped.

Moreover, I wouldn't even use $169, because if you just follow the 'good deals' board, you can always buy 2TB disks for $139.

 

2) Again, you chose to leave electricity cost out of the equasion, and yet compared scenarios with different number of disks.

If you insist on leaving electricity out, then you should populate the box with the same number of disks.

 

So, when taking 1 and 2 above into consideration, here is the slightly revised arithmetic:

 

a) 5 x 1.5TB disks:

(6000 GB * 0.928)/($800 + $69 + (5 * $109.99)) = 3.9240 GB/$

 

b) 5 x 2TB disks:

(8000 GB * 0.928)/($800 + $69 + (5 * $139.99)) = 4.7318 GB/$

 

So now it turns out that that with the 2TB disks you'll be getting 20.6% more GigaBytes for your dollar.

( 3.9240 * 120.6% = 4.7318 )

 

Of course, the above calculation is not quite precise either, because that particular box can be obtained for much less than what you assumed.

But that doesn't matter for what I am trying to say here:

The main point is that when you talk about 'sweet spots' you have to take into consideration a lot more things than just dividing disk gigabyte by disk dollar.

Purko

 

In response:

1) I always just use whatever prices are available today.  If I could buy a 2 TB WD Green for $139 today, then I would have used that figure.  It doesn't make much sense to me to compare what we expect to be a reasonable sales price when we have no way of knowing when that price will be available again.

 

2) Yes, I purposefully left electricity out of the equation again because I felt that I adequately demonstrated that the cost of electricity to run one or two extra drives is negligible (on the scale of less than $10 per year even in the most extreme circumstances, see this post).  It seems to me the most fair way to make the comparison is to keep the server's actual storage capacity constant, as opposed to the number of disks.  After all, that's what the user really cares about, how much data they can store.

 

I agree that if (good quality) 2 TB drives can be had for $139.99, then they are obviously the better deal.  However, that currently isn't the case.  The particular calculations above are just examples using today's prices; any numbers can be subbed in there to reflect the current market prices.  My purpose is to discuss and refine the method, not to gain actual results.

[WeeboTech]

2) Yes, I purposefully left electricity out of the equation again because I felt that I adequately demonstrated that the cost of electricity to run one or two extra drives is negligible (on the scale of less than $10 per year even in the most extreme circumstances, see this post).  It seems to me the most fair way to make the comparison is to keep the server's actual storage capacity constant, as opposed to the number of disks.  After all, that's what the user really cares about, how much data they can store.

 

I think credibility is lost with this.

Electricity IS a factor over a long haul. Add in the fact that it is always going up.

The amount may seem negligible, but that may be for one point in time.

More drives, more electricity, more future maintenance, higher exponential costs.

 

I think this becomes more important as you reach "critical drive mass" heh new term.

Where your server is so full of drives you cannot add any more.

 

When the cost of electricity jumps up a bit. I've seen a real bite in my wallet from this.

 

I think a key factor is jumping on prices when they are attractive.

 

A few dollars for drives in a specific time point is one thing.

When you factor in rising power costs, then heat, airflow, cooling. There are other parts to this equation.

[purko]

Rajahal, in that example above, the box has 5 disk slots.

Five is your limit after which you have to think about another server.

Yet, you insist of crunching figures with only 4 x 2TB disks, versus 5 x 1.5TB disks.

I guess you can prove whatever you set out to prove.  I give up.

 

And another thing, nobody around here buys disks on whatever todays price is.

There are ALWAYS sales going on.  Especially on newegg.

People buy disks whenever a decent one is on good sale.  Which is like always.

 

To tell you the truth, I am really lost about what you are trying to accomplish.

In a free market there are gazillion things being sold at gazillion different prices.

And each of those is good for somebody for some particular purpose.

There is no 'one-size-fits-all' formula.  If you think there is, well, good luck finding it.

 

I'm done here.   

 

 

[NAS]

Sales must me an American thing as disks rarely go on special in the EU.

 

Also most of the calculations seems to be based on the assumption you need all the free space now. Each disk gets cheaper as time passes and you would be folly to buy loads of spare HDD space that you wont need for months. You need some sort of sliding scale where disk price comes down and capcity comes up ultimately leading to a new sweet spot.

[neilt0]

Sales must me an American thing as disks rarely go on special in the EU.

 

Not true. This is a particularly good deal at the moment (there are a few left): http://www.hotukdeals.com/item/600745/2tb-hdd-for-less-than-100-inc-vat-d

 

2TB for £84.99 ex. VAT.

[Chris Pollard]

Sales must me an American thing as disks rarely go on special in the EU.

 

Not true. This is a particularly good deal at the moment (there are a few left): http://www.hotukdeals.com/item/600745/2tb-hdd-for-less-than-100-inc-vat-d

 

2TB for £84.99 ex. VAT.

 

Yup Aria have the same price too, free shipping if you sign up to their forums

 

http://www.aria.co.uk/SuperSpecials/Other+products/2+Terabyte+%282TB%29+Hitachi+7K2000%2C+32MB+SATA+?productId=38391

 

[NAS]

Hmm perhaps your right. I have WD blinkers now.

[bubbaQ]

No one has mentioned total cost of ownership (TCO), annualized costs, or reliability factors.

 

For example, when your drives get old, they are more likely to fail.... you need to have a replacement schedule for drives, and check SMART values and replace them when they get in the 30K hours lifetime range.  In other words, drives have a finite lifetime.

 

Power does cost and needs to be included, and allocated over the LIFE of the drive.... not annually.  But PSU will have the most impact on power, and after that mobo and CPU.  Drive power comes in close to last for most users.

 

Every component has a lifetime.  Its TCO has to be calculated over its lifetime, and then the annual costs compared.