Data Center Strategies

Back To The Future (2) – Oracle + Sun

In my previous post on this topic I promised to take a look at how a more “vertically integrated” set of IT suppliers might change things. In this post we’ll look at the challenges faced by a combination of Oracle and Sun.

The first challenge that any company will face is that a modern IT organization is orders of magnitude more complex and diverse than in era when IBM ruled the data center. For example, most IT organizations today run tens or even hundreds of applications from a wide variety of vendors on a diverse set hardware. That’s a far cry from the IBM era when an organization might have one or two computers and a handful of applications. So the big question is whether true vertical integration all the way up the hardware and software stack for the entire enterprise is even possible today. I’m inclined to think not, because even Oracle only has a handful of the applications needed by a modern enterprise

So if full vertical integration isn’t an option, the key questions are “how far up the stack can Oracle go?” and “does the convenience of vertical integration for part of the IT environment offset the issues of vendor lock-in and higher upfront purchase costs?”

The answer to the first question is that Oracle could offer a well integrated stack for Oracle databases and any applications from Oracle that live on top of that Oracle database. Using Sun’s hardware and operating system (along with OEM products like the Hitachi Data Systems enterprise arrays) they have access to a complete suite of enterprise hardware and system software. The problem is that much of that hardware is not widely used in the enterprise and many IT organizations have long since made the choice of another supplier, so Oracle has to persuade them to rip-n-replace a significant chunk of their infrastructure in order to benefit from integrated Oracle solutions.

The answer to the second question will depend a great deal on the previous choices made by an IT organization. For example an IT organization running Oracle on Sun hardware using Sun storage products may find the idea of tighter integration quite attractive. On the other hand an organization using HP servers and EMC storage to underpin it’s Oracle applications may not be interested at all.

An area of particular concern is the hypervisor, given Oracle’s recent acquisitions, it’s clear that they want to be a player in the server virtualization world. The problem is that many large enterprises are already deploying VMware ESX and to a lesser degree Microsoft HyperV and Citrix XenServer along with a suite of tools to manage the virtual environment. If Oracle fails to displace these products and becomes a “virtualization island” with it’s own set of management tools, then much of the vertical integration advantage disappears, as IT organizations are forced to work with two sets of tools, one for Oracle solutions and one for everything else.

Posted by: Nik Simpson

Cisco Adds Rack Mount Servers

When Cisco launched it’s UCS blade system in March, they were adamant that they did not plan to get into the rack mount server business. As I pointed out at the time (see here) the lack of rack mount servers would be a significant barrier to entry in the Enterprise because:

1. Many customers are still wary of the vendor lock-in, limited scalability, and pricey nature of blades and continue to buy rack mount servers in large quantities.
2. IT organizations try to keep the number of hardware companies they deal with to a minimum in order to maximize their buying power and simplify support for the server infrastructure. Telling companies to buy blades from Cisco and rack mounts from another supplier is not likely to fly with many customers.

So guess what, today Cisco announced a range of 1U and 2U rack mount servers designed to complement the USC blade systems. I suspect that this is the result of early feedback from potential resellers and end users, i.e. “if you want to be taken seriously, then you have to bring more than blades to the data center gunfight.” Given the short timeframe between this launch and the March UCS announcements, it’s hard to avoid the conclusion that Cisco always had a contingency plan here, and just hoped they wouldn’t need it.

The range of servers is still limited compared to the major server vendors, but 1U and 2U dual processor systems are certainly the sweet spot of the market, and should help the get a foot in the door at many enterprise IT organizations. Of course, this is all moot until Cisco actually has general availability on USC and its rack mount cousins, at the moment they seem to be as rare as rocking horse droppings outside of Cisco data centers!

Posted by: Nik Simpson

Back to The Future

As you may have gathered by now (see [1],[2]), I’m skeptical about Oracle’s plans to create a one stop shopping experience for IT organizations by combining their software offerings with Sun’s hardware. But let’s assume for the moment that I’m wrong and Oracle really does plan to go down that path.

Such a move would attempt to turn the industry clock back 40 years or more to an era when IT companies were vertically integrated and IT organizations sourced pretty much everything from one mainframe vendor. At that time there was very little in the way of an Independent Software Vendor (ISV) community, you bought your hardware, operating system and applications from the same place. Over time vertical integration gave way to the horizontal integration that we have today, were companies pick and choose hardware, operating systems, and applications from a wide variety of vendors.

Each model has distinct advantages and disadvantages:

• Vertical integration: Buying everything from one company as an integrated package should certainly be easier to manage and gives IT organization “one throat to choke” when things go wrong. On the other, it’s likely to be more expensive to purchase because the IT organization is heavily committed to a single vendor’s solution which weakens their bargaining power.
• Horizontal integration: The large and competitive selection of vendors for every aspect of a horizontally integrated IT infrastructure means that prices are low and there is considerable scope for “let’s make a deal”. On the other hand, a horizontally integrated IT infrastructure will require more on site integration and ongoing management attention.

If Oracle is right and the industry is ripe for a move back to vertical integration, then we have to ask the question “how will this impact existing vendors.” That’s a big subject and too much to cover in one blog, so I’ll tackle in a series of blogs over the next weeks. The vendors I plan to look at are:

• Oracle + Sun
• Cisco + EMC + VMware
• Dell
• HP
• Microsoft
• IBM

Posted by: Nik Simpson

Epiphany on The Road to Santa Clara?

Ever since the Oracle/Sun deal was announced (see my comments here), the rumor mill has been working fulltime on the issue of what Oracle will do with Sun’s Hardware business. Last week, in an interview here, Larry Ellison tried to put rumors of the imminent demise of Sun’s hardware business to rest. The gist of the interview is Oracle is going to embrace Sun’s hardware business with the zeal of a convert.

Unfortunately, the effect of these protestations of undying hardware love were somewhat undermined this week when Sun made the details of it’s courtship of Oracle public (see here for the gory SEC details). Apparently, all Oracle really wanted was part of Sun’s software business and a minority stake in the company, apparently Sun turned that idea down.

Plan B was to go in with an unnamed hardware company (HP according to most sources) with Oracle taking the software and HP getting the hardware business. That plan failed as well, leaving Oracle with two choices, walk away, or buy Sun with its hardware baggage and all, and sort it out later.

Apparently buying the entire company was preferable to walking away. Subsequently, on the road to Santa Clara to finalize the deal, Larry had a religious experience and became a born again hardware geek.

Of course, you cynics out there might be thinking that Oracle had to say nice things about the hardware business, at least for now. The last thing they want is to confirm a Sun customer’s worst nightmare and SPARC a mass exodus.

Posted by: Nik Simpson

EU Slaps Intel with €1.06 billion ($1.45 USB) Fine

By now, I guess everyone has seen the incredible news that the EU has levied a €1.06 ($1.45USB) fine on Intel for "illegal sales practices". As you can imagine, Intel was unhappy about the news, while AMD applauded the move.

From a technical analysis perspective, there are no immediate implications. Intel will still continue to ship it's entire portfolio of processor products, including the new Nahalem (XEON 5500 series) processor. Well, at least for now. However, one has to question the long term implications, including whether or not Intel has the cash to sustain these product lines given the fine. Intel, like many other companies in this economic climate, are watching every dime. Will this fine affect their ability to produce product in sufficient quantities to meet IHV demand? These types of issues make IHVs (e.g. Dell, HP, and IBM) very nervous. Often times, there simply aren't enough chips to go around in the server market. That's why every IHV scrambles to be "on stage" with Intel when they announce a new processor...to guarantee they are "first to market" and corner as much of Intel's capacity as possible. The more processors the IHVs have on hand, the more serves they can manufacture... and the easier it is to keep the competition from shipping similar servers. It's been this way in the server market for years. How many times have customers attempted to purchase a server from an IHV, only to find out that the ship date is extended 3+ months?

Another question: will this judgment hinder Intel from making more investments in future products? Chips design is a long process. New processors are on the drawing board 5+ years in advance. If Intel is forced to make hard decisions, R&D may be the place.

Part of what we've been blogging on here in DCS is the increasing competition within the IT industry...including Ciscos's entry in the blade server market with UCS and Oracle's Virtual Iron acquisition. We think it's an inevitability that larger companies will begin to compete across technical, geographical, and vertical market boundaries. Companies that were once "close partners" are now suspiciously watching each other from across the table. Stands to reason the the governing agencies will be closely following these unfolding events.

[posted by: Drue Reeves]

What Rubbish!

This morning, the web is alive with stories claiming that security researchers have found an unfixable hole in Microsoft’s Windows 7 operating system (and by implication the closely related Windows 2008 R2.) Needless to say this makes for good headlines and plenty of web traffic for any site that puts up the article. But the fact is, the “hole” is not a Windows 7 hack at all. The hack (if you can call it that) involves getting physical access to the machine, changing the boot order to boot from something other than the boot drive (i.e. a USB key) and then loading code into memory that takes over the boot process before the real operating system loads.

Anybody with even a modicum of savvy can see that this is not a security hole in the operating system, since it takes place before the operating system is even loaded. While the code they’ve written targets Windows 7 (much better chance of headlines) the concept will work with any operating system on Intel hardware. If you can take control of the hardware before the operating system is loaded in memory and change the OS image in memory, there’s not much the OS can do about it. So the defense is one that is already well understood by IT, physical security, if you don’t have it, then all your other security measures are meaningless.

Posted by: Nik Simpson

Xeon 5500 VMmark Numbers – Reading the Tea Leaves

The introduction of Intel’s Xeon 5500 combined with this week’s announcement of VMware ESX 4.0 has produced a number of new VMmark benchmark results.  Here’s a summary of the results from HP, Dell, and Cisco:

	Result	Configuration
Cisco B200-M1 (Blade)	24.14 @ 17 tiles	2 x Intel Xeon 5570 (2.93 GHz) 96 GB RAM (DDR3 @ 1066MHz) VMware ESX 4.0
Dell PowerEdge r710 (rack mount)	24.00 @ 17 tiles	2 x Intel Xeon 5570 (2.93 GHz) 96 GB RAM (DDR3 @ 1066MHz) VMware ESX 4.0
Dell PowerEdge r710 (rack mount)	23.55 @ 16 tiles	2 x Intel Xeon 5570 (2.93 GHz) 96 GB RAM (DDR3 @ 1066MHz) VMware ESX 4.0 (pre-release)
Dell PowerEdge M610 (Blade)	23.90 @ 17 tiles	2 x Intel Xeon 5570 (2.93 GHz) 96 GB RAM (DDR3 @ 1066MHz) VMware ESX 4.0
HP ProLiant DL370 G6 (rack mount)	23.96 @ 16 tiles	2x Intel Xeon 5580 (3.2 GHz) 96 GB RAM (DDR3 @ 1066MHz) VMware ESX 4.0 (pre-release)
Higher scores in the benchmark indicate higher performance, you can find a full explanation of the benchmark here.

Although these results are all very similar, there are some interesting points to note:

1. HP's result is the slowest but uses the fastest processors. I suspect this is an anomaly caused by using a pre-release build of ESX 4.0. This suspicion is based on the two Dell PE r710 results where the only difference is the change from a pre-release build (23.55 @ 16 tiles) to a production build (24.00 @ 17 tiles). I strongly suspect that the HP ProLiant result on the ESX 4.0 release build will be up at the top given the faster processors used in the configuration.
   
2. All the configurations tested used 96 GB of memory, despite the fact that all the systems (with the exception of the Dell PE M610) could support substantially more (as much as 384 GB on the Cisco B200.)  This strongly suggests that 96 GB is the sweet spot for memory when configuring these systems for server virtualization. If more memory would have produced a higher benchmark result, then it’s hard to understand why nobody bothered to try.
   
3. There appears to be little to choose on performance, for example the Cisco blade has a 0.58% performance advantage over the PowerEdge 710, which could easily be accounted for by differences in the storage configuration. So customers should focus on value-add for features that simplify deployment of server virtualization and the vendor’s ability to provide professional services for virtualization implementations.
   
4. Prior to the release of the Xeon 5500, all VMmark results were completed using ESX 3.5, which makes it impossible to know how much of the performance boost comes from the Xeon 5500 and how much comes from ESX 4.0. So be careful comparing ESX 3.5 results with ESX 4.0 results.

One thing that is clear is that for 2-socket servers, Xeon 5500 with ESX 4.0 is king of VMmark hill by a wide margin. The best AMD-based result with ESX 3.5 is an HP ProLiant DL 385 G5 scoring just 11.28 @ 8 tiles. I suspect we’ll have to wait a month or two for AMD’s Istanbul processor before we get some comparable results on ESX 4.0.

Posted by: Nik Simpson

There was a House in Santa Clara They Called the Rising Sun

So the soap opera we’ve been following for the last month has come to it’s conclusion, with a shocking twist in the final episode. After being spurned by the big blue Knight from Armonk,our heroine, Sun Microsystems, was rescued at the last minute by Oracle, presumably riding on in on white charger. The $9.50/share price is at the the top of the range for the rumored IBM deal, so it’s at least some good news for shareholders.

For Sun’s IT customers it’s good news/bad news. The good news is that Sun will not got nova for the time being, a prospect that was becoming all too plausible in recent months. In the short/medium term (I’m guessing the next 3-5 years) Oracle will continue to service and support UltraSPARC/Oracle customers, to do otherwise would be suicidal.

The bad news is the uncertainty this creates around the long term future for Sun’s hardware products, particularly its UltraSPARC high-end servers. The question is, “What does Oracle plan to do with the hardware business?”, because lets face it, the reason for the acquisition is Java, not hardware. Oracle has seen the future of hardware for databases, and its a commodity-based scale-out cluster; not large, proprietary, and expensive servers. So I believe it’s a question of “when” not “if” Oracle decides to stop further development of Sun’s high-end server line.

For the commodity server products the future doesn’t look too rosy either. Sun’s market share for commodity servers is negligible compared to IBM/HP/Dell which leaves them with two problems that they need to address:

1. There isn’t enough business in just selling hardware for database clusters to support a product line, so Oracle has to decide whether it really wants to be in the hardware business at all.
2. Going into accounts where Oracle is running on HP/IBM/Dell hardware and trying to kick out the incumbent hardware vendor is going be a hard sell and will do nothing to endear Oracle to their former hardware partners.

So you have to wonder why Oracle would choose to spend R&D dollars on x86 servers to go after a small segment of the commodity hardware business and turn their erstwhile partners in the hardware space into competitors.

Posted by: Nik Simpson

IBM and Sun: On Again, Off Again Relationship

Well, you had to see this one coming. The NY Times is reporting that IBM has now withdrawn its $7 Billion acquisition offer for Sun Microsystems. This is after IBM lowered its offer from $10 per share to $9.40. Sun, reportedly was "OK" with the deal as long as IBM would make guarantees not to walk away from the deal. Why would Sun want a guarantee that IBM would not walk away from the deal? Because of exactly what has happened. Think about it...how does the market view Sun now? How do customers view Sun now? A.M. Sacconaghi (interviewed in the NY Times article) hit the nail on the head... "For I.B.M., given its size, this was never a transformational deal, but in Sun’s case, it’s an extremely material event. This leaves Sun in a tough situation,Sun was on a path to selling itself, and this will inevitably raise questions in customers’ minds, no matter what Sun says, about its commitment to a go-it-alone strategy.”

Personally, I don't think we've heard the end of this deal. IBM really doesn't want Sun to fall into HP, Dell, or Cisco's hands. Sun really doesn't want to go-it-alone until another suitor arrives.

Last time I posted a blog on this deal, I described what this meant for other server vendors in this market. Given this situation, let's do that again.

• Hewlett Packard:  There's not a lot for HP to be gained by buying Sun. HP's server line is superior. It's storage line is competitive to Sun. And, HP's software division is the management leader. HP isn't really into applications the way Sun is, so taking on Java would be a stretch. BUT -- HP could play a little chess. HP could feign interest in Sun, thereby driving Sun's price back up, and scaring IBM that they perhaps made the wrong move. IBM might then come back to the table and Sun would be off the market again. HP would make IBM spend a little more cash than they wanted. In the server market, it's dog-eat-dog. If you can get your competitor to spend a few extra bucks so that they can't use it against you, you do it.
• Dell: <sigh>....For Dell, it doesn't get to look any better than this. If IBM and Sun are truly on the outs, and Sun's perception in the market is headed downward, then now is the time for Dell to get Sun. This is now the 3rd time I've blogged about this [1,2]. Dell needs to stop moving down market into PCs and gadgets and start thinking enterprise. The margins are better, the customers have deeper pockets, and Dell's brand needs a credibility boost. Dell, do yourself a favor and throw your hat into the ring.
• Cisco: I would say "See Dell" and be done, but Cisco actually has even more to gain than Dell. Dell, compared to Cisco, has a much more well-rounded enterprise portfolio. Cisco is in need of rack servers, storage, server customers, service people with server knowledge, etc, etc, etc. Cisco can't afford to build their server product portfolio organically. That would take too long. Cisco simply can't compete in many enterprise accounts without a more well-rounded portfolio. Will enterprise customers who need rack servers, storage, service, and blades buy the blades from Cisco and everything else from other vendors? No. Cisco's strength is its brand. Scooping up Sun would prove to the IT industry that Cisco is a player to be reckoned with and not a flash-in-the-pan.
• Fujitsu: Fujitsu gets a reprieve from having to buy SPARC from IBM. Plus Sun is now their new best friend since Sun will need the revenue. Fujitsu needs to put their SPARC exit strategy into motion. Fujitsu, meet Intel and AMD, they have nice chips.
• AMD and Intel: These two shouldn't get too excited about Sun coming back. Its only a matter of time before something happens. AMD, just like Fujitsu, time to figure out that Sun exit strategy.

[posted by: Drue Reeves]

Xeon 5500 – Rack Server Announcements

With the exception of Sun (who have delayed their announcement) all the major x86 server vendors launched a slew of new rack servers based on the Xeon 5500 processor announced yesterday (see here). Before I start, there’s two points to get out of the way:

1. I’m not going to go into great detail about each new configuration, there’s plenty of coverage elsewhere, not to mention the vendor’s respective websites for details of feeds and speeds.
2. I’ll cover blade announcements in separate blog.

Instead I’ll focus on the highlights and compare some of these machines with the models with existing systems. First, let’s look at what Dell announced, the PowerEdge r710 which compares most directly to the existing AMD-based PowerEdge r805. The key differences are are shown in the following table:

	PE r805	PE r710	Comments
Memory	128 GB DDR2	144 GB DDR3	12.5% capacity + bandwidth improvements
PCIe slots	3 x8 slots (Gen 1) 1 x4 slot  (Gen 1)	2 x8 slots (Gen 2) 2 x4 slots (Gen 2)	Same number of slots, but available I/O bandwidth is ~60% greater

The performance difference on the VMmark benchmarks speaks for itself with r710 (23.55 @ 16 tiles) roughly doubling the performance of the r805 (11.23 @ 8 tiles).

HP went much further than Dell, introducing replacements for their Proliant G5 systems across the board (including DL360, DL370, and DL380). The G6 variants offer roughly double the memory and massive improvements in I/O bandwidth vs. systems they replace. For example here’s a quick comparison of the DL360 G5 and G6 models.

	DL 380 G5	DL 380 G6	Comments
Memory	64 GB DDR2	144 GB DDR3	225% capacity + bandwidth improvements
PCIe slots	2 x8 slots (Gen 1)  2 x4 slot  (Gen 1)	2 x8 slots (Gen 2)  4 x4 slots (Gen 2)	50% more slots, but available I/O bandwidth is ~300% greater

To put the performance difference in perspective, HP has published a TPC/C result on the DL360 G6 which is roughly 50% higher than any previous 2-socket TPC/C number (631,766 tpm vs. 404,462 tpm). In fact the DL 360 G6 result is only marginally slower on TPC/C than the DL 580 G5 4-socket which achieved 639,253 tpm as recently as January of 2009.

Like Dell, IBM only announced a single new rack mount server, the x3650 M2, which shows similar improvements over it’s predecessor the x3650.

The key point for anybody considering dual-socket server purchases in Q2 or later is that they should wait for availability of the new Xeon 5500 products which offer considerably better “bang for the buck” in terms of both performance and energy efficiency. The only caveats relate to the use of these servers in virtualization clusters:

• Don’t mix AMD-based and Intel-based servers in the same cluster because although the server virtualization extensions are similar, they are mutually incompatible when it comes to live migration of virtual machines from one server to another.
• Try not to mix old and new Xeon-based servers in the same cluster because it will negate some of the virtualization extensions in Xeon 5500 that are not supported on older Xeon processors.

Posted by: Nik Simpson