The Storage Economist

We have all heard it before.

“I can go down to <insert name of local retail tech store> and buy a 1TB disk for $60. Why is your enterprise disk so expensive??”

We have all tried to fumble our way around price and cost and how the enterprise disk is so much better… yada yada yada. Well, someone went down to their local retail store, used petty cash and built a system. Yes, the unit cost of disk was cheap ($160), but at the end the total cost was not at all cheap — $8,500 for that TB of disk.

A colleague of mine in Australia did this experiment. I asked him to document the process, and the math behind the amazing results. I have done some editing to his material (they speak a different English down-under), but I give him full credit for the write-up and the total experiment.

Remember that Price ≠ Cost.  Happy reading.

True <discount store> Disk TCO

Consider that if a person was to buy a 1TB drive from <discount store> for $168 (that was the <discount store> price as of January 2011 for a half decent-quality drive, even though you can pay as little and $79), they would need to buy a second drive to have redundancy like we deliver, and they would need to buy a third drive to have hot-sparing like we deliver. That makes $504.

Now consider that the formatted size would actually be 940GB not 1TB so the cost is closer to $534 per Terabyte usable.  A 1TB USB drive to use at home seems to be what uneducated (ignorant?) people always use as a comparison so you can see already their argument is flawed.

Then factor in how you would back that up. The current technology capable of backing that up in a single run is LTO4 and a cheap but reliable LTO4 drive costs $2,800 with one year’s warranty. You would need a minimum of 41 tapes to maintain the GFS (grandfather, father, son) daily-fortnightly-twice annual backup rotation and retention for one year that we use, at a cost of $69 per tape ($2,829).

So, the cheap 1TB of storage at <discount store> is now looking like $504 + $2,800 + $2,829 = $6,133.

And no one is looking after it, and monitoring it. No one will come out and replace/fix it if it has an issue at 2:00 a.m. on a Thursday (best you might get is 1 years return to shop warranty). The alternative would be to buy 41 extra hard drives to make copies on to, at $168 each although you could then drop the hot spare so only 40 extra drives would be required to use more <discount store>  disks as the backup target. 1.

The 41 extra hard drives is to counter the argument that you just need to make a copy of the data on another disk the way people would in their own homes. If you do that you don’t actually have a point in time backup. You just have a single copy of the current data for DR that does not meet any regulatory requirements. If you were to make a copy every day and every week as you would with a backup regime, once the 1TB SATA drive was more than half full you would need 41 extra drives to be able to have point in time recovery for every business day for the most recent 14 days, and every fortnight going back a full year.

Now realize that this is only a SATA disk and that is in fact the slowest type of storage we offer and has no cache in front of it. It would not be capable of supporting databases or enterprise applications and has a very slow access rate because it is only USB2.0 or at best Fire Wire.

NOTE:Most enterprise storage arrays have a background process that does block level error checking so for a more accurate comparison you need to factor in the time it would take to write a script that did a file-by-file CRC check and then you would need to manually monitor the results and do the restores for all the corrupt content. You would need to do it at least twice a week so that is a minimum of 52 weeks times twice a week times 30 minutes each time at an IT Support Tech pay rate, which is $2,300 a year.

So now your 1TB <discount store> drive is costing you close to $8,500 a year to deliver something close to enterprise levels of service but still delivering woeful performance — with no on-site or after hours support.

This is all to the extreme and of course if you bought 100 drives from <discount store> you would demand a substantial discount but the TCO has more backup and management costs than capex spend. If you could buy a tape drive with 3 years warranty the TCO would be less than $2,800 per year and a proportion of the backup media could be recycled for a couple of years so the media cost would be less over a multiyear model. BUT you normally only get a 1 year warranty on a hard drive from <discount store>, so the cost modeling over a single year is still valid.

Remember, if you build all systems with cheap SATA internal or DAS disk there are issues backing them up. Anything larger than a few hundred GB can’t be reliably written out to tape every night (like a database) without dedicating a backup device to that specific system. Now we have a $10,000 tape library attached to every server. This is not very cost effective or space efficient. If we just use a single standalone drive for every system instead of a library, there is no bar-coding to manage the thousands of media required each year in an enterprise environment. Now tapes are getting lost, misplaced or simply overwritten because no one has a central console to track and monitor them.

Bear in mind that the reason SAN even exists is because having disk specifically locked in a given server is inefficient. Not all boxes will use up their storage at the same rate. But extra space in one server can’t be used by another server the way SAN space can be assigned to whichever system needs it.

People who have been in IT long enough remember the 80s and 90s and it wasn’t great for data management. The models from back then simply cannot scale to the size required in the 21st century.

So there you go. It is possible to construct an apples-to-apples comparison from discount store disk drives to enterprise disk architectures. There are features and functions that chance the price and total cost:

• Performance
• Drive failure protection with RAID
• Reliability of the device
• Disaster protection
• Data protection (backups and replication)
• High performance connection (Fibre channel, not USB ports)
• Management and operations

In discussion total price with total cost, you may want to point your management team to a paper outlining all the types of costs that make up storage TCO.

Good selling!

Tweet

Posted in CAPEX, ROI, Storage Economics

Comments (7 )

Post Comment

It seems that my references to **discount stores** was lost in the HTML-translations…ignore that previous post.

Although I think some will be tempted to reply that you’ve convoluted their point - that drive-for-drive, the price of a home-use 1TB drive is but a fraction of the “enterprise ready” ones we sell, and that by adding in all the extras you’ve tried to hide that fact…(wait for it)…
.
I agree with you.
.
In fact, you could add in a few more things…
.
* That **discount store** drive doesn’t offer Thin Provisioning, nor do any of the standard operating systems you could use it with. You have to assign 100% of the capacity to the host/server, even though you KNOW it isn’t using it all…with an nitelligent storage array, that unused space can be shared/utilized by other hosts, driving down your actual $GB *USED* (as opposed to “usable” or $/GB raw).
.
* Want to put the busiest part of your capacity in Flash, and the least-used parts on SATA so you get the best $/IOP and $/GB? No can do with that store-bought drive…but your enterprise array can do this for you - along with using Very Large DRAM cache to handle read prefetch AND to delay/defer writes. With the store-bought drive, you’ll never exceed the raw performance of the drive - when used in any of the enterprise-class, large-cache, automated-tiering-enabled arrays, you will get performance than cannot be matched by ANY drive you can buy down at the **discount store**.
.
That inexpensive drive also doesn’t provide remote replication, point-in-time snapshots (as you mentioned), nor any form of data reduction (compression or deduplication), nor data encryption (you CAN buy drives with encryption, but at a premium price), or data corruption protection (T10-DIF, RAID parity checking, drive scrubbing, etc.).
.
And we could go on…
.
The challenge we face as vendors in this market is that we (collectively) monetize the value-add of our storage arrays in the price of our drives (flash+fibre/SAS+SATA/fat SAS)…justifiably, we will argue, because our arrays provide levels and quality of service that the naked drives alone cannot.
.
One oft-overlooked fact is that the drives we use aren’t the same ones you can buy at the **discount stores** . Even the SATA drives we use are “enterprise-class” - hardened to handle the heavy vibrations typically of dense packaging we deploy (remember the video of the Sun Thunder that suffered a performance hit when you YELLED! at it?). The motors, lubricants and bushings are better on the drives we use, and they often have larger caches and more intelligent command queueing … all these add to the price.
.
Thanks for broaching the subject…

Hi David

This is just plain amazing ! Excellent blog !

I like wjhat you’re trying to say and I think it’s imoprtant to explain to consumers of storage that there are costs which we take great pains to conceal from them to manage their data. However, I think your discount store comparison is rather extreme. Using that comparison, weakens the argument because it focusses the debate on spurious objections, such as the lack of warranty that comes with an additional $15 discount. The debate is not really about that $15. Sure you can save a little some money and add a lot of risk, just as a high end storage vendor can help you spend a lot more money to eliminate a little more risk. But a more realistic scenario for most folks is something closer to the middle.

Most technologies have a cost curve that shows very low costs for very low performance, and very high costs for very high performance. But it’s not linear in between, it’s exponential. Meaning that the best tradeoff lies somewhere in the middle. Depending on your motivation (performance or cost), comparing a technology at one end of that curve to something at the other end will justify your preference. But in defending your choice that way, you conveniently omit the solutions in the middle of the curve with typically better tradeoffs.

I give you this amateur/enthusiast example as an alternative. Consider that I buy 3 x 2TB hard drives at $250 each (with a 3 yr return to store warranty), plus a 4-bay eSATA RAID enclosure for $250. Formatted, I’ll have 1.8GB of mirrored disk space with a spare hot-swappable drive. I don’t need a 24 hr engineer, in the unlikely event that a disk fails, I’ll swap it myself and replace the failed drive on warranty in the morning. To back that up, I’ll take a full image backup of my data to a 3rd 2TB disk once per week, and a daily incremental to a 4th 2TB external disk. That’s another $500 for backups. I’ll dedicate a further couple of external 2TB hard drives to a duplicate backup process giving me a copy I can take offsite (although more likely I’ll look to put my offsite copies in the cloud.) Finally, I’ll run my CRC checks on the backup drives immediately after the backup completes to assure their quality (and that of the source) so as not to impact the availability and performance of my primary storage. Over the 3 years I’ll run this configuration, I’ll spend less than $500/TB per year.

With a configuration like this, I’ll sleep pretty well at night. Granted I could spend more to add more performance, redundancy, monitoring, support etc. But the return on any further investment diminishes rapidly and I’ll need to decide if I can afford the luxury of sleeping on the feathers of baby geese, while being fanned by angels.

Sure you can probably pick holes in my technical design and point out additional risks I haven’t mentioned, just as there will always be scenarios in a top end solution that don’t work out as the provider intended. But the point is the real challenge you need to address is not the end-user advocated discount store solution at the other end of the cost curve, it’s the amateur/enthusiast/IT solution in the middle.

For a more sophisticated analysis, one should create the curve not by plotting a single measure of absolute performance against the purchase price of each option. Instead one should calculate a personalised composite value for various aspects of performance and plot that against the annual price (taking into account, for example, the warranty period of your technology). It may be more accurate to use cost (or TCO) for post spend analysis, but in making a purchasing decision one should use price, and not include some costs that legitimately belong in a TCO such as the notional costs of resources that are already committed and cannot be avoided, e.g. floorspace, if you already have the space and no means to avoid paying for it, you should not include it in your price.

[...] couple of comments from my previous post, in which I presented a TCO analysis of the retail-store-hard-disk-as enterprise-disk, have driven [...]

[...] blogged about cheap disk a few months back, and you can see these entries here and here. I am starting to see a divergence in thoughts and plans around storage intelligence and [...]

[...] Why again? Well, it was about a year and a half ago when last blogged, and much has changed since then, although the subject is still front and center. David Merrill has certainly discussed this numerous times as well, including this always-amusing post from March. [...]

[...] assets, but positions you for sustainable growth into the future. See what Claus Mikkelsen and David Merrill have to say about storage efficiencies. [...]

Post a Comment