Storage TCO: Intangible Cost Factors
When comparing enterprise storage systems with regard to cost, it’s only human to dwell on the easier, concrete metrics, like box price, features, and capacity. But there are other, less measurable factors sailing in under the radar that may leave you with a system that’s cheap on paper but expensive in practice.
And here are those factors, in increasing order of measuring difficulty: Time, Effective Capacity, and Responsiveness.
Time. The time a storage administrator team spends maintaining and tweaking a system for required performance and service levels is an expense commonly acknowledged but hard to factor in. You may not consider it a component of system cost since you pay your staff’s salaries no matter what tasks they carry out or how long the tasks take.
But is this true? If it takes hours to provision new volumes, resize existing volumes, make backups, and perform other management tasks, then the associated people resources are unavailable for assignments that could otherwise contribute to enterprise profitability. A system with high demands on administrator time is a costly system.
Effective Capacity. TBs are easy to price, but let’s look more critically at your organization’s capacity demands. Is that targeted 150 TB really necessary or is it actually to over-compensate for a capacity-guzzling system and to accommodate traditional storage management practices? Are you using or contemplating a system that will require “excess” capacity that might not be used for years – to cover for uncertainty due to scaling limitations? Will the system cause you to order new capacity when, in fact, pockets of unused capacity go untouched because they require too much administrator time to reclaim?
Another thought to consider: Is your capacity target, in actuality, inflated to compensate for features that miss the mark, such as capacity-consuming full copies rather than differential snapshots or effective thin provisioning? A system that guzzles capacity or requires the IT organization to overload on capacity is a costly system.
Responsiveness. Try to put a price on the cost to your organization of slowed storage services. In the best of all worlds, the IT center is the ultimate enabler, not hindrance, letting business activities flow and allowing every department to get its work done. When comparing systems for cost, take the time to consider what mechanisms the system offers to help administrators optimize responsiveness. For example, how long does the system take to provision a new volume? Does it provide extra snapshots so your development organization can do more testing? Does it provide flexible and frequent enough logical backups to ensure that you can roll back the data in case of a logical error? Do such options involve so much storage overhead as to potentially limit your ability to even offer them?
Consider, also, the system’s lead time in responding to user requests. For example, how long does it take the system to resize a volume? Imagine a different reality, in which a storage administrator – or even more junior IT team member – can provision storage at will, in a data-safe, virtualized environment.
If these activities will take too long – and how much time is “too long” is a fuzzy metric to assess for cost but not insignificant at all – then that system may literally slow down your organization, causing individuals and departments to be less efficient and perhaps even miss business opportunities. In short, that’s a system that may cost the company money. This is pretty ironic, given that if there’s any one industry that’s added speed – and at lowered costs – to the 21st century business environment, it’s IT. A system that hampers IT responsiveness is a costly system.
In short, how much are these intangible “assets” worth to you? If they’re missing in a system, how much might their absence cost your department and even the enterprise?
Avoid the trap: When comparing two storage systems, don’t fall prey to evaluating only the metrics that stare you in the face and measure easily. Do all you can to explore the less tangible cost factors. The results may surprise you.
We know we promised some time back to drill down into SATA vs FC reliability … stay tuned, we have that one queued up for two weeks from now…
Not much to disagree with here; welcome to the thin provisioining dynamically sized volume party, late though you may be.
But this I really don’t understand.
Is your capacity target, in actuality, inflated to compensate for features that miss the mark, such as capacity-consuming full copies
Given that the XIV mirrors data for reliability, that’s a bit of an odd comment. A 180TB XIV is around 73TB usable. Is the XIV built on a feature that misses the mark, I wonder? Certainly, the XIV is “a system that guzzles capacity” like no other I’ve ever come across, thinly provisioned or otherwise.
Also consider, it’s easy to get lost in the weeds when you start analyzing raw storage versus useable storage versus how much storage do you really need to run the business. Given the many options on the market today, I think the only clear answer is to evaluate your actual and projected cash outlay for storage over a three and five year term, then define soft costs as a separate category. Simply put, how many hard dollars does your business spend each year and expect to spend in the out years to acquire, support, power and manage storage? Assuming you can provide equal or better reliability, performance, environmentals and functionality then it should come down to a cost comparison. Do I really need to know if the power in my 120V home outlet is coming from a Nuclear, Oil or Coal fired power plant? I just want clean, uninterrupted power at the lowest cost. Based on the TCO model I developed I can’t find any solution from EMC, HDS (Sun/HP) that competes with XIV in the 60TB+ range. In one scenario comparing 240TB useable of DMX4 to 360TB’s useable of XIV, I found that even if EMC gave the hardware and software away for free it would still cost more for EMC over a five year term!
J.R You need to ensure your TCO is comparing apples for apples. Against DMX you certainly aren’t.
If you look at the modular arrays from any of the above vendors, load them up with SATA then you’ll find them price competitive and generally more capable.
Time
Ease of management varies but given most can provide much better scalability than XIV, then you don’t need to break out into managing two or three seperate systems.
Go for a system with vrtualisation features and once deployed they’re just as simple to manage.
Effective Capacity
All have more efficient Raid options providing much more usable capacity per $ and subsequently much higher potential scalability.
All utlise modular building blocks, you can buy what you need now, afterall disk capacities increase and costs per GB reduce as a direct result.
All have thin Provioning in one form or another. Some have space reclamation (LUN Shrink) and drive spin down.
All have space efficient snaps as well as regular clones and a few other options not available with XIV.
Most if not all can resize volumes on the fly , again the virtual system can take seconds to minutes to achieve this, simply request a larger capacity.
With any of these you don’t need to deploy 180 disks now. Why would you, all consume datacentre floor space, power and cooling as well as support costs
Responsiveness.
Provisioning time varies, but the virtual boxes are just as fast as XIV, seconds to minutes for volume provisioning.
Snap shots / clones / snapback etc. XIV may have an edge in numbers supported, but given the ease of management message, that goes out the window once you start having to manage hundreds of snapshots. Also where’s the application integration for snapshots ? VSS hardware provider alone doesn’t cut it.
Assuming I run out of capacity on one of the above modular systems, I know there’s generally a short lead time on enclosures and disks. Equally I can purchase just what I need, how long is the procurement process for an XIV frame ?
These aren’t one trick, one size fits all solutions, you have many more options open for the future. If you can truly predict your 3 to 5 year storage requirement with any real accuracy and are willing to buy yourself into a capacity and performance dead end. Then good luck, because you’ll probably need it.
Sorry but I’m just not getting the value prop.
Dear John:
I read your comment “If you look at the modular arrays from any of the above vendors, load them up with SATA then you’ll find them price competitive and generally more capable”.
I bet there might be a case for price competitiveness IF and only IF such vendors would load them up with SATA drives only… yet, they don’t, why, because they can’t, they are incapable. Why they can’t? Because traditional RAID technologies simply would not be able to provide the right balance of performance and reliability for tier1/2 customers. Otherwise, why wouldn’t they give me a fully loaded SATA DMX or USP V today?
The reality is that there’s no apples to apples comparison at all. XIV is simply oranges… how about turning a DMX into an orange and see how it fares in multiple dimensions? (cost, ease of use, time to deploy, support, responsiveness to changes, power, space and performance… using only SATA drives). I doubt that would ever happen.
Please don’t turn this into another FUD junkie. One is enough.
F.S.A
Biased comment; vendors DO provide modular arrays with dual parity RAID for reliability and efficent capacity use, along with high performance, all based on SATA. NetApp (for whom I work) for one.
Plus there’s thin provisioing, dynamic volume shrink/grow, deduplication on primary and not just secondary, NAS on the same box, SATA, etc, while still bowling along at a more than respectable rate.
There’s a tendency to look to the DMX and USP in comaprison with the XIV. That I can’t understand. These boxes have their place, but the XIV isn’t in that space at all. It’s a monolithic lump with an incredibly strange value proposition; a SAN that’s being positioned and compared here for Tier1, when IBM claim “consolidated utility storage for fast growing, dynamic mixed, and emerging workloads”. That is, not SAN.
I can’t square that circle.
“I bet there might be a case for price competitiveness IF and only IF such vendors would load them up with SATA drives only… yet, they don’t, why, because they can’t, they are incapable.”
Really that’s news to me, go speak to SUN, HP, HDS or even IBM, all can do this as can EMC to a lesser extent.
“Why they can’t? Because traditional RAID technologies simply would not be able to provide the right balance of performance and reliability for tier1/2 customers. Otherwise, why wouldn’t they give me a fully loaded SATA DMX or USP V today?”
I wouldn’t class DMX or USP as mid tier storage. It’s most definately Tier 1 and isn’t what you should be comparing XIV against unless you’re trying to skew the playing field.
“The reality is that there’s no apples to apples comparison at all. XIV is simply oranges… how about turning a DMX into an orange and see how it fares in multiple dimensions? (cost, ease of use, time to deploy, support, responsiveness to changes, power, space and performance… using only SATA drives). I doubt that would ever happen.”
Again, DMX isn’t your competition, no serious storage admin would put XIV in the same category as DMX or USP. XIV at best is mid range tier 2 aka EMC-CX, HDS-AMS, HP-EVA, SUN, 3PAR etc bordering on tier 3, only it’s not really scalable enough for archive or cloud.
It’s SATA only.
Doesn’t scale to enterprise levels.
Few enterprise features.
Lacks broad O/S support.
No mainframe.
Completely unproven.
As I said comparisons with DMX are not really relevant, if you wnat to comapre with a mid range array with SATA then I’ll be happy to oblige.
I could easily do a comparison against one of the above mid tier arrays and come away price competitive, Look at Fishworks from SUN, or maybe EVA from HP, both are an absolute doddle to setup and manage and both support enterprise SATA/FATA. Also both have much more scalablity and many more features available than XIV. Both support lower power enclosures, since they don’t ship a motherboard, processor and memory with every module. They both support other raid variants allowing for better capacity efficiencies and fewer drives, consuming less power and both systems manage and scale SATA/FATA in exactly the same manner as they would Tier 1 FC disks. The performance argument is spurious at best and the ease of management angle doesn’t stand up against these arrays.
As far as I’m concerned the information being provided thus far is pure marketing. Lets face facts it’s a majority node cluster in a rack, coupled with 1Gb backend switching and all backed by UPS to protect a volatile writeback cache. Yes there’s 8Gb of cache per module, but remember some of this is taken by an O/S (Linux), a large amount will be dedicated to block mapping, some will be policy memory for system use, snap tracking etc, the remainder will probably be a 50/50 R/W cache split with the write back cache halved again due to mirroring. So what’s the real figure ? it doesn’t look like a global cache to me, which is what USP and DMX have. In fact if lefthand did FC rather than just iSCSI I would be comparing XIV with their architecture, it’s so similar just missing FC and the UPS.
Please don’t turn this into another FUD junkie. One is enough.
No FUD here, I’m asking questions and voicing my opinion based on experience in this marketplace and on these technologies. I just don’t see how you can get so excited about a one trick pony.
“Really that’s news to me, go speak to SUN, HP, HDS or even IBM, all can do this as can EMC to a lesser extent.”
I think you missed my point. I didn’t mean that technically, because I know they can. The question is would they ever sell a fully loaded DMX SATA-only array? And yes, I mean for tier1.
“I wouldn’t class DMX or USP as mid tier storage. It’s most definately Tier 1 and isn’t what you should be comparing XIV against unless you’re trying to skew the playing field.”
Actually, I never positioned DMX or USP as mid-tier. I’m truly comparing XIV to them because I see the value proposition at a much much lesser price tag than any of those two. No skewing, just thinking outside the box.
I see you and I have fundamental differences about it. I mean no disrespect but I sense you are trying to explain and reject something you haven’t configured, seen and deployed yet. So how should anybody actually believe you as much as they should believe me? I think that will ultimately be driven by the market.
Honestly no ! I don’t think any vendor in their right mind would sell all SATA for tier 1, without at the very least a second synchronous replicated array, and that remote copy just covers the data availability aspect, not the potential performance implications.
Is that what IBM are actually proposing here ? That XIV is capable of replacing Tier 1 storage in the datacentre. That customers should just throw out the years and years of engineering know how and proven, tried and tested data availablity and recovery features ?
BTW your correct I haven’t deployed, configured or used the XIV…not many have. However even if I had, I don’t think that would place me in a position to make such a recommendation, given the limited architectural information available, the products limited field deployment and so far unproven track record.
Well good luck with the market, but I honestly think Tier 1 territory is a much tougher and unforgiving place than you imagine, and that unless IBM have something up their sleeve they were sold a lemon here. But assuming XIV is tier 1, where does that leave the IBM DS8000 ? where’s it’s value prop ?
Starting to loose the value in why IBM sold my CIO a DS8300 with SVC nodes in front of it. Perhaps XIV is going to decomission both those products?
I truly want to become more educated on XIV so thank you for these posts.
I have a question about your Effective Capacity TCO point - esp around “pockets of unused capacity”
I agree this is a HUGE problem in the industry.
But what I didn’t see in your post was how XIV fixes this problem, unlike other products in the market. Is it differential snapshots and thin provisioning?
But you also claim extra snapshots for app testing. But what happens to those extra snaps when testing is over? You also mirror your data for reliability as well, right?
I recommend storage products for optimization, so pockets of unused capacity is right up my alley. I am trying to make the XIV link to optimized storage.
You might like reading about 33 different types of costs that make up stoage TCO. Your comments are correct, but there are factors of hard costs and soft costs that are defined by the end-user or cash-counter…..
Dave
http://www.hds.com/assets/pdf/33-types-of-costs-for-storage-tco.pdf
[...] came across this recent article from ThinkStorage on the less tangible types of TCO costs. They have clearly outlined 3 or 4 of the 33 types of costs [...]
There seems to be some emotional attachment to storage architectures that have been working now for many decades. XIV is approaching the Teir 1 market from a fresh direction and actually doing ok (a direction you may argue that they weren’t first to think of but they were first to get working in a Teir 1 market). I don’t think we should get too luddite about this technology and give it a go. Remember it has got the brains behind it that the Symm did back in the day so it isn’t bereft of an understanding of enterprise storage.
I have been involved in an XIV deal. When the customer got the boxes installed was chuffed to bits with ease of use and the performance (and before you ask they where not a mid teir usage customer!!)
Summary - chill out and give it a go!