Saturday, May 19, 2007

Those who ignore the past...

One of the good things about being an Old Storage Geezer (OSG) is that you develop perspective. After you've seen enough real upheavals, every silly thing that comes along doesn't seem like such a crisis. If you're a lucky OSG, then your hard won perspective allows you to see a foggy bit forward as well as backward with clarity.

Right now, I am becoming fascinated with a new foggy notion. Like a song that I can't get out of my head - maybe its the lyme's, or Denny's Mad Cow - but I keep seeing visions of a quantum shift occuring in our industry. The catalyst is Nand. If these visions are right, a fundamental shift in value is likely to occur, and many of the existing industry leaders are likely to be fading memories in just a few years.

Let me put a few things in perspective for the various insects out there that will poo-poo (i can't beleive i just used that expression) my ideas, or perhaps my earned right to have any ideas about this industry at all.

Contrary to rumor, I'm really not that old in dog years. I'm about the same age as Mark, Howard, Jon, Robin, John McArthur, etc - so I guess if I am an OSG then we are all getting to be OSGers.

Old or not, I've been around long enough to have personally experienced several quantum technology shifts in this industry - and I've earned the right to expect some respect (grudging or otherwise) from those coming after.

I once made my living by being one of the best head alignment specialists in the Northeast - for 300MB Control Data Disk Drives. I bet most of you have never seen one of these - they were about the size of dishwasher and had a dozen 19" platters. We cleaned the heads with alcohol and IBM punch cards and aligned them to the servo track by hand. It was kind of fun, unless you flinched and crashed the heads. I sneezed once...omigod...that was quite a meltdown...

Then, along came Winchester and poof (i can't beleive I used that expression either) there went the osciloscope and the t-bar...and the Control Data Corporation...

A few years later, along came 51/4" disks. I was at the plant when we closed Digital's 9" winchester disk factory in Kaufbeuren, Germany. The looks on the faces of our employees - who never saw it coming, and did not understand why - stay with me even now.

Later, I led a team of great people who introduced the first hetrogeneous RAID system. Lot's of people said it couldnt be done, customers wanted to buy storage from their system vendor, customer wanted to buy storage optimized for a specific server, etc, etc. But before we knew it a whole industry sprouted up around us - proprietary disk arrays, and many companies that made them, vanished.

Then, a few years later, I woke up one night in a hotel in Colorado Springs and sketched out a diagram for a network storage architecture on the hotel stationary. I showed it to people the next day and told them this was going to be the future of storage. Most of them laughed. That diagram - sketched in pencil in 1996 - became ENSA - the Enterprise Network Storage Architecture. And eventually direct attached disk systems, and many of the companies that made them vanished...

Sometimes the attenuation on my 'future vision' is off by a few years. In 2000, we launched the concept of Managed Storage Services. The wild idea that you should let the experts take care of your storage. Too early. Only a few brave souls, like my buddies at Arsenal Digital, dug down and stayed with it - but now the idea makes perfect sense to thousands of customers, big and small alike. And, with the buzz around Google and S3, Peter and Bill's orginal SSP concept is back on center stage as well.

And of course, CDP...but that story comes in July...

The point is that some of us OSGs actually can see above the clutter - can sense patterns emerging, and if you are smart enough, and lose the arrogance of thinking you're right all the time, you might actually learn a thing or two - young whipper snapppers.

Now back to the point...the next big thing is SSD...and it is going to kick your ever loving a$$ if you don't figure it out quick.

8 comments:

Robert Pearson said...

How would you Implement this?
Just replace all the "rotating rust" with SSD in the ENSA?
Just a brief 100,000 foot overview is sufficient.

I posted a comment about RAIDFlash (slow SSD) at:
http://www.drunkendata.com/?p=1095
It has not generated any interest.

I always liked the ENSA concept as a Strategy. The problem was I fell in love with Gene Amdahl's "concept" Storage box back in the 1997-99 time frame and have been trying to convince someone to build it ever since. It moved the ENSA inside.

We still have the "innie/outie" debate. The rule seems to be that "smart" and "very smart" boxen are slow relative to "very dumb" and "dumb" boxen. I think that "slow" and "fast" are in the eye of the Storage beholder.
The new twist on this seems to be to have light speed fast, but very dumb, boxen operating under a very smart virtual layer. Lots of metadata flying around out there.

The real way to do this is starting to come. That way is to incorporate the metadata into the Information entity or object, there is a difference.

RE: "Peter and Bill's original SSP concept"
There was never anything wrong with the SSP, ASP or MSP concepts. Their demise was two-fold. None of those companies were ever started to last for the long-haul and customers were reluctant to buy in to the concepts. Nothing wrong with either of those goals if you were one of the "new millionaires" or a guy who kept his job rather than take a chance on a new concept.

Online Storage is having the same slow going today. It is only being successful with Web businesses. Traditional businesses are having the same problem with Online Storage they had with SSP, ASP, MSP.

You may ignore this comment since I am not associated with any TLAs.
Writing these comments is like virtue. Virtue is its own reward and that's a darn good thing.

Robert Pearson said...

I submitted this comment several weeks ago to Hu Yoshida's post of:
"Going Beyond Virtualization"
http://blogs.hds.com/hu/2007/05/going_beyond_virtualization.html

It has never been posted.
What do you think?
Good or garbage?

Thanks you for the “new trajectory” concept.
I have long been thinking Information needs to be defined as "vector" or "scalar" based on a number of factors. Your “new trajectory” makes this at least possible with HDS.

In your statement:
"With a new trajectory there are so many new facets for performance that there is no way to provide a comparison with other vendor’s products. The closest comparison that we can provide is in the maximum IOPs that a system can generate."

IOPs are scalar information as are all of the other "performance" facets.

I use "information" for scalar and "Information" for vector. The distinction is a definition in evolution.

I wait with great anticipation to hear the "vector" properties that give a "new trajectory". These "vector" properties could finally enable the much needed SFO (Search, Find, Obtain) capability.

This should be the "Dawning of a new Age" for Storage.

Kirby said...

Hey Robert, thanks for the notes.

the new SSD stuff is plug compatible with HDD (2.5" and 3.5", SATA or FC, etc). Its 10-50 faster on random reads, uses less than half the power, generates no heat, and no vibration. Prices are falling 80% a year. capacities over 100GB are on the near term horizon. But there's more to it - at 50x you dont need as many drives to meet performance, you don't need controller cache, you dont need expensive packaging, power, and cooling...in fact, you need to rearchitect the whole array infrastructure.

No you dont replace all the rust - but the rust drives have to get way bigger - way, way bigger, and you need to be able to move data in different media (HDD and SDD) media teirs within the array very efficiently.

the first vendors to figure this out have an opporutnity to tranform the value hierarchy in the market.

forgive me- but I havent had my coffee yet, or the lyme is still slowing my brain - I am having a devil of a time figuring out your second post - can you translate to pigeon english for me?

Robert Pearson said...

Thanks for the reply.
From your reply it appears you are talking "real" SSD and I am talking strictly Flash. I would hope that "real" SSD would drop in price to Flash levels but I don't expect it to. I agree fully with the "real" SSD as plug-in replacements for "rotating rust", as the TCO differential decreases and where it adds an advantage.

I wonder who those vendors might be that are perceptive and flexible enough to seize the "opportunity to transform the value hierarchy in the market"?

Robert Pearson said...

RE:"can you translate to pigeon english for me?"

Thanks for your reply and candor.
I am not sure my translation will be useful since I was looking for help trying to grasp this concept myself.

Here is where I am coming from:
Two things-
1.) CAS (Content Addressed Storage) or any system to bring organization to ad hoc Information spaces.
2.) Dion Hinchcliffe's using Web 2.0 and SOA to do this ad hoc Information organization:
"Eleven Emerging Ideas for SOA Architects in 2007"
http://hinchcliffe.org/default.aspx
[post excerpt]
"3. Deeply embracing URI addressability. Of all the things in this list, this might be the most important one. The hyperlink is the fundamental unit of thought on the Web and it should be in your service designs and (hopefully granular) schemas as well. Giving each discrete piece of information, every service, and all content a globally addressable URI instantly gives a service, and the data it carries back and forth across its interface, access to countless new consumption and reuse scenarios. The most important of these is the leveraging of network effects via -- often social -- link propagation along with the ability to make all URI addressed information potentially crawlable, thereby making it transparent via search. The possibility of letting people find your service via an intranet or Web search engine because of the great content it has might seem a little odd at first but then again, that's what makes things work so well on the Web. You can learn about URIs on Wikipedia, and they can be a SOA's best friend."

You can use URIs and CAS to create link maps (or hyperlinks) in application space. This is a huge maintenance nightmare.

It would be better to have the ability to have direct hyperlinks into Storage. Doing this has been discussed for years.

I am talking strictly ad hoc Information space here.

With the hyperlink how do you tell the importance of the hyperlink or Information it points to?
In a vector hyperlink the importance would be included and be intuitive or obvious to the destination. It may be possible to do this with SOA.

Nigel said...

Hi Kirby,

Really interesting post. Do you know of anyone who is looking into SSD seriously?

I posted in a similar vain over at RM (http://blogs.rupturedmonkey.com/?p=49)
a while ago talking about SSD and Texas Memory Systems. It seems from a performance point of view its a no brainer. But the green issue is becoming more and more important these days, and as has been mentioned SSD requires far less power and generates little heat.

Nigel

PS. Im looking forward to my next birthday as Im hoping the age I hit my get me a little more respect ;-) Often older = wiser ;-)

Rob Peglar said...

Kirby et al,

Just ran into this blog. Interesting. Nigel, to answer your question, we've had DRAM-based SSD implemented for over a year now. Works great, lasts long time. Put it side-by-each (as they say up north) to the 15K drives; clustered, virtual, immediately available, iSCSI or FC, the whole shebang. Performance is, well, screaming.

Kirby, nice to hear from you. See you around the circuit.

Rob

IDCStorageHW said...

First of all, great discussion. I direct storage research at IDC (http://www.idc.com/getdoc.jsp?containerId=PRF000326) and completely agree with the prospects for SSDs. 2 years ago, we saw the underpinnings for NAND flash begin to come together, potentially giving HDDs a run for their money. IDC complimented its existing semiconductor research, and now has a dedicated SSD analyst and some of the largest SSD OEMs as clients.

One only has to look at the evolution of the Apple iPod to see the potential for NAND flash. There are three primary variables to the equation for determining what type of storage to use: Capacity, performance, and cost. In other words, choose any application, weigh capacity and performance requirements against the projected cost of the storage, and choose your poison.

Alas, it isn't quite that easy. Performance is not just about I/Os per second or read/write speeds. It is also about reliability, power efficiency, and robustness – all under the umbrella of performance. Moreover, cost isn't just about the initial purchase price. It is also about the cost to cool, power, replace, protect – in other words, TOC.

In short, we have a solid (pun intended) team of analysts looking at SSDs from a semiconductor perspective to an end-solution perspective, analyzing both supply-side and demand-side perspectives.

The punch line for SSDs?
In the enterprise think Tier 0 (zero).
In Consumer electronics, think good enough.
In PCs, change your definition of a PC.