EMC 2009 Strategic Forum -- Storage

Continuing on a previous post from our recent event, I thought I'd use this post to recap the main points that Joe Tucci shared with the audience as part of his executive keynote. 

Joe Tucci just hit a few highlights in his session -- spending far more time on introducing private cloud concepts (along with Paul Maritz). 

The bulk of the storage-related strategy detail was handled by Dave Donatelli in a private breakout session.

But I think it's useful to cover how Joe teed it up, and the key points he emphasized from the podium.

Introducing Information Infrastructure .. Again!

Joe Tucci spent some time touring through the highlights of "what's hot" in EMC's information infrastructure portfolio, exclusive of VMware which was covered subsequently by Paul Maritz.

The first stop was EMC's heritage and core strength -- storage. 

Given the audience, it was pretty well understood that EMC was a leader in the market and was doing well -- they all follow the numbers pretty closely, so we didn't have to make a case that we knew what we were doing in this category!

Joe chose to focus on a couple of areas he thought was important.  One of earlier slides was simply a discussion of current popular topics in the marketplace. 

Storage consolidation and increased utilization is even more popular than in the past -- no surprise.  Use less, use it better.

What most people don't realize is that there's renewed interest in business continuity these days. 

Joe didn't say it, but many of us believe that this is a direct consequence of widespread virtualization, and VMware products such as SRM that dramatically lower the cost and complexity of many business continuity scenarios.

Joe's key point -- and one we've all made before -- is that there are a lot of smaller companies who try to build a product around a single capabilitiy, e.g. dedupe, etc. 

What customers want is a broad set of product platform capabilities that include these features, as well as many others.

Now, he didn't say that "tape was dead" -- nobody really believes that -- but customers were certainly investing in a whole lot less of it lately, and tape was being pushed farther and farther down in the hierarchy.

When it came to data deduplication, he stated that EMC was now well positioned to play in all the different datadeuplication scenarios, including recently primary storage (think NetApp)

And then Joe used his podium to take a little wind out of Data Domain's sails, so to speak.  In Q4, EMC did over $90+ million in backup-related dedupe products, which would technically make EMC the market leader in the category from a revenue perspective. 

He also stated that the ramp had just started, if I remember correctly.

The "Other Dedupe" -- Flash

Joe doesn't describe it this way, but many of us do (Chad Sakac started it!) where we have fallen into the habit of describing SSD / flash / enterprise flash drives as "I/O dedupe".

The context is pretty simple. 

Disks supply two primary metrics: capacity and IOPs. 

Traditional dedupe lets you use fewer disks to deliver a given capacity. 

Enterprise flash allows you to use fewer disks to deliver a given IOPs level -- hence "I/O dedupe".

Joe was very clear -- he sees this as the most important advancement in the storage industry ever.  No fundamental change has occured in storage media since IBM introduced a 5MB rotating rust drive way back in 1956. 

I think he ran out of words trying to explain just how big this was.

He then made some interesting points that made the crowd sit up and take notice.

First, he highlighted just how quickly the prices were coming down on EFDs (enterprise flash drives), showing the slide going from 40x FC drive prices to only 8x FC drive prices in about 12 months. 

He flatly stated that he expected this price decline to continue apace in coming months, and EMC was committed to driving the price down for this important new technology.

He also shared that EMC completely sold out of EFDs for DMX and CX in recent quarters.  We were increasing supplies, but customers had figured out just how powerful this technology was.

Now, if you're reading The Storage Anarchist, you might know that not every storage vendor sees things the same way :-)

Joe then presented an interesting example as to just why EMC customers loved this stuff so much. 

He did a side-by-side comparison of a current DMX 4 built two ways -- one built the traditional way (244 300GB 15K FC drives -- a reasonably high performance configuration), and one built with a different mix of technologies: 8 x 73GB flash, 136 300GB 15K FC drives, and 32 x 1TB SATA drives.

The before and after numbers speak for themselves: 60% more IOPs, 17% lower cost -- in addition to things like 28% fewer drives and 21% less power and cooling. 

What I got out of this -- in addition to Joe's comments -- is that a little flash goes a very long way, especially in larger configurations such as shown here.  The corrollary might be that smaller configs may not see quite the impact shown here. 

Joe also stated that making these devices work well in existing customer environments took a lot of work.  Sure, we're starting to see a few vendors bring their initial offerings to market, but these competitive platforms were not purpose-built to handle this technology (as is the case with the current DMX and CX), nor did they appear to have the degree of integration and optimization.

Given EMC's significant lead here, it makes you wonder what we'll do next? :-)

Virtualization Changes Everything ... Again!

And, to close off this post, I'll share with you the chart that Joe used to show how well EMC was doing in VMware environments.  He put up a side-by-side showing EMC's overall "branded market share" according to IDC (smaller numbers), and then showed an average of three separate surveys done by analysts of storage market share in VMware environments (the really big circle).

Paul Martiz was on hand to unequivocally state that EMC had achieved this entirely on their own; VMware had provided the exact same opportunities to each of its storage partners, it's just that EMC had capitalized better than most.

Which is exactly as it should be, no?

Comments

[Jerry Thornton]

Maybe instead of I/O dedupe it should be called spindle dedupe because you're not removing or reducing I/O, you're removing unnecessary spindles to support a given I/O. Maybe I'm being pendantic, but just my two cents.

[Chuck Hollis]

Good thought!

[Andrew MacDonald]

I/O Dedupe is what we do in PowerPath - great in VMware environments. We do spindle and I/O dedupe, talk about comprehensive solutions..

[Bob Primmer]

Good post, Chuck.

[StorageTax]

The DMX4 config comparisson looks interesting, but I cant help but think an array configured as described has a very narrow band advantage over a "traditional" configuration.

Less power and cooling is always a good thing. The real question I have is what is the real performance benefit of the hybrid config ? I find it difficult believe that any array is going to deliver 60% better overall performance by converting less than 1% of your useable storage to flash, especially when you convert nearly 50% to slower SATA drives.

I would love to see EMC publish some test data to back up these numbers with real workloads. No single application numbers but real honest to goodness multi-application mixed workloads, 20+ hosts connected to the array.

Im guessing that there will actually be a small performance degradation, simply because in the real world 90% of work isnt done on 1% of data.

Well thats the way it is in my world anyway.

[Chuck Hollis]

Hi StorageTax

Good news! Just ask any EMC "SPEED" Guru (a large team of field performance experts), and they have access to literally dozens of real-world application profiles, both done in our labs and from customer environments, with both large single hosts (e.g. mainframes) as well as multi-host environments.

I think you'll be pleasantly surprised.

The magic is that most "hot" applications have pronounced disk hot spots that are easily revealed with just a bit of analysis.

On a DMX, the large non-volatile write cache does a good job of soaking up most writes, but sooner or later has to dump to disk -- EFDs can help there. And random reads tend to defeat cache algorithms, but they make EFDs shine.

There are a number of before-and-after comparisons floating around that have somehow escaped our corporate perimeter, if you go looking. All are pretty amazing.

It'd be great to be able to have a detailed discussion with you, or whoever, and understand your environment better.

Thanks for the comment!

-- Chuck

[the storage anarchist]

StorageTax -

Indeed, not all environments fit into the 80/20 (or 90/10) rule, where a small subset of spindles support the overwhelming majority of the IO workload.

But most do.

This is why wide-striping works, for example: every spindle supports a small subset of the total workload. The idea is similar with Flash Drives - put the most used data on the fast storage to deliver the IOPS, the least-used on slow SATA, and demote everything else based on utilization (heck, wide-stripe everything else, if you'd like).

The EFD advantage is not just IOPS, though. EFDs can deliver response times unattainable with wide-striping. So rather than a degredation, most will see a rather significant improvement in performance.

Admittedly, we're at the beginning of leveraging flash, but saving money for the most performance-hungry applications is an excellent place to start!

[Martin G]

Should you not wide-stripe everything anyway Anarchist? Just have separate wide-striped pools? One for flash, one for FC and one for SATA?

StorageTax - we've had some modelling done looking at upgrading our DMX4s; a relatively small amount of flash would make a significant difference and this is for a mixed workload.