Killing Me Softly

Early this week an IBM blogger mentioned that IBM had increased the density of storage on tape by some impressive factor of 40x or thereabouts that would result in tape cartridges of 35 TB. As a research exercise or proof of concept or some such thing.

Consider me unimpressed. Well, no, not quite. I am impressed in the same way that I might be if I took a trip to the Ripley's Believe It Or Not exhibit at the circus. Weird. Bizarre. Freaky. But not terribly useful to the real world.

I also had two more rational reactions.

The first of these was that the "announcement" reminded me a lot of a briefing that I attended many years ago. The briefing was given by Sun, and it was all about their new multi-core, multi-threaded processors.This was a big thing at the time: Sun had been selling some really really big systems (E10000s and E15000s for those of you with long memories). And they were really really expensive. Yes they occasionally went out the door at 90% discounts. But often these systems were $10,000,000 or so. Just for the system. No storage. No apps. No database. Which seems pretty absurd by today's standards, doesn't it?

Anyways, sales were way down, and more importantly, Sun realized their chips had pretty much hit a wall. It wasn't going to be possible to keep making them faster, not with the existing chip architecture, and not at a reasonable cost point, given the volume of chips they were making.

So some engineer had a really good idea: ditch the old stuff, and come out with new processors that were inherently multi-threaded and multi-core. They were going to be cheap. And they were going to be fast. And they had a really good roadmap. But as I listened to the presenter, I had a "wait a minute" moment. Here were the facts:

• New processors to be 8-32 times as fast as the "old" processors on which the big systems were based
• Roadmap to go to 64 times faster and beyond
• Speed not just "theoretical" gee whiz, but they believed that it was reflective of real world workloads, particularly in some core areas in which Sun excelled, like Web servers and applications
• System price would drop accordingly (because I now need just 1-2 processors, instead of 64)

So. If I take systems that were selling for $1,000,00 to $10,000,000, and if I replace them with systems that sell for $50,000 but are capable in many cases of running the same workload, I have two choices if I want to stay in business. Choice one: sell 20 to 40 times as many systems as I was selling previously. Choice two: charge a high price.

But I was blown away at the time by the fact that nobody in Sun seemed to have made this connection. You cannot replace a very expensive product with a very inexpensive product and sell the same number of units (or even double unit shipments) and make the same amount of money.

The strategy made no business sense at all.

It was a recipe for turning a billion dollar revenue stream into one that was worth a hundred million. At best.

Even more shockingly, there was no sense that the decline in revenue was going to have secondary impacts, such as dramatically reducing the ability to fund research, maintain headcount, fund customer service, and so on. (As it happens, the "plan" was to offset or grow revenue by giving away stuff--read free software licensed under the GPL--which wasn't very useful in terms of replacing lost revenue either. But my point is not to deconstruct the Sun implosion.)

Now what impresses me most is that their is a near perfect parallel between what Sun did with processors and what IBM is doing with tape: capacities to increase by a huge multiple. But I would be willing to bet my last dollar that there will not be any similar increase in cost or in units shipped to offset this. No tape cartridge is going to cost $2000 (roughly 40x what a current LTO cartridge costs). And they sure aren't going to sell 40x as may of them.

Which means that tape revenue is about to undergo a fairly precipitous decline.

A terminal decline, perhaps.

Because there is another side to this coin. Which is the second of my more rational reactions. And the other side is this: the capacity will only matter to people that want capacity and won't matter at all to people that care about performance.

Let me put that another way: tape capacity only matters for archive. Because tape performance has long since passed the point where it matters for backup.

As a thought exercise: would it matter if this tape drive, with a capacity of 35 TB per cartridge, had a performance of 40x that of LTO4? (That would be about 5 GB per second.)

No. Not even a little bit. In fact, increased tape performance hasn't mattered since about 50 MB/s. Everything after that, every tape drive faster than that, has been largely useless from a performance perspective. Because nobody can get the data to the drive fast enough. (And this is one of the many factors that has lead to the success of disk backup with deduplication over the last few years: it is not just that disk can run at 1,000 MB/s or whatever, it is that it can do that with 20 streams or 220 streams simultaneously. We don't care if a stream is slow or fast.)

As evidence of this, ask yourself how many tape drives somebody has, and how much data they back up. Multiply the theoretical speed of their tape drives by the number of drives, and how much does that come to? I can think of one organization with 200 LTO3 and LTO4 drives, and an 8 hour backup window. That would give them the ability to back up 800 TB, give or take, in their backup window. They have about 200 TB to back up in a full backup, once per week, and less than 100 TB to back up every night. So for them, tape speed doesn't matter at all. Even if they had a bunch of IBM's theoretical 35 TB, 5 GB/s tape drives, they would still need 200 of them.

What will this tape drive and cartridge be good for? Archive. Long term retention. And for this, it is really good because I don't care that it will take 3 days to fill a cartridge, I just care that I will only need a very few tape drives and a very few cartridges to archive me entire data set.

Which means that a customer with 200 tape drives probably now needs just half a dozen tape drives, if they backup to disk with deduplication first. That same customer, with space for some 50,000 plus tape cartridges in libraries on their data centre floor (at the cost of several thousand square feet of floor space, power, and cooling) will now require but a few hundred cartridges at most.

So if I need just 5% of the resources I had before to do a job, is that good? Yes.

But thank goodness I am not a tape vendor. Because it sounds like my revenue is about to take a pretty massive 95% reduction.

And for those of you that say tape isn't dead: if the revenue for tape declines by 90% in the next few years, does that mean tape is alive? Or just irrelevant? (And remember that in the world of IT and technology, irrelevant equals dead.)

Tape may not be dead today. Right now. But it is as thoroughly and as completely doomed as a technology as Sun was (is?) as a company. Drastically downsized after 75% head count reductions and precipitous declines in revenue, many parts of Sun are well on their way to irrelevancy.

At least tape vendors (or vendors' tape divisions) will have good company.

Comments

[Alex Chekholko]

I don't think you're saying "tape is dead", I think you're saying "the tape industry won't grow". Tape is the optimal solution for some problems, e.g. the "archive" problem that you mentioned. And that's not going away.

I agree that it's not a growth industry, but I plan to still be using LTO-X tapes in a decade...

[Daniel S.]

This just doesn't make sense to me. Why does EMC and other vendors keep offering bigger and bigger drives in their arrays or implementing deduplication technologies? By your logic you're making it cheaper to store the data because you don't need such huge arrays anymore. Datadomain and other deduplication technologies would in fact be signing their own death warrant because nobody would need to buy disk anymore.

In reality you need continued capacity growth of tape media for the same reasons you need continued growth of disk technology... we have an ever expanding need to store more and more data. If you can sell the expanded technology for cheaper you're individual unit sales may not be as much but you'll likely sell more and remain competitive when other vendors can provide increasingly smaller and cheaper systems that outperform the big ones.

[Ricardo Costa]

Daniel S., it starts to make sense when you keep in mind that bigger tapes will be extremely important to archive data, not for backup.

For backup, bigger disks (along with faster appliances) will be extremely important to backup data, not to archive it.

That said, both parties (tape and disk vendors) are signing their own death warrant. It's good, natural, and expected when talking about IT and our personal lives.

Everybody needs to reinvent and adapt to new times and needs. Tape, as their business is today, will die. Disks, the way it is today, will also die somehow and sometime. I can't see any surprises here...

Are we still seeing k7 walkmans around? Phones that only talk (funny)? Cars becoming more oil consuming? Let's face it: everything changed, died, and those companies are still there reinventing themselves (some way stronger them before - see Apple, Intel, Oracle, etc).

What prevents this scenario to make sense is to look into new directions but keeping our minds in the past. I understood what Scott tried to position and I agree. I also understand your confusion, but you are holding yourself in a "words game" rather then have a broader vision about those changes. If you keep your attention to the analogies and "wording", you will miss the whole thing around this article and IBM's announcement.