Eric Stahl's Homepage
Eric Stahl was Senior Director of Investor Relations at BEA. After
joining the company in 1999 he spent his first five years focused on the
development of the application server market and drove the marketing
plan for WebLogic Server.
WebLogic Server (9.2) is back on top with the latest SPECjAppServer2004 benchmark result. This configuration achieved 7,174.56 SPECjAppServer2004 JOPS@Standard. You can see all of the configuration details here. Sun did some comparisons to the Oracle/HP and IBM numbers here.
SPECjAppServer is a trademark of the Standard Performance Evaluation Corp. (SPEC). Competitive numbers shown reflect results published on www.spec.org as of January 22nd, 2007.
It's been a while since my last SPECjAppServer2004 update so here we go again...
After a relatively quiet summer seven benchmarks were posted in August, one in September and two so far in November, bringing the total number of comparable results to 25.
In August IBM showcased high end and low end configurations. The high end was a twenty node cluster, each with two 3.6 MHz Xeon processors (40 cores) running WebSphere 6.1 on SUSE Linux Enterprise Server 9. The total throughput was 4,368.02 SPECjAppServer2004 JOPS@Standard. The low end submission was a single node app server with one Power5+ dual core processor running WebSphere 6.1 on IBM AIX 5L V5.3 producing 349.11 SPECjAppServer2004 JOPS@Standard. These two submissions briefly held the highest throughput and JOPS/core numbers.
In August HP took back the high end lead with a four node, 32 core Itanium2 system running WebLogic Server 9.1 on HP-UX 11i version 2 that put out 4,915.49 SPECjAppServer2004 JOPS@Standard.
In September Sun struck back with a ten node cluster, each with 4 dual core Opteron 885 processors (80 cores total) running WebLogic Server 9.1on Solaris 10 that put out 6,662.98 SPECjAppServer2004 JOPS@Standard.
This week a new score was published. It is the first configuration submitted by Inspur, a server manufacturer from China.
Here are the configuration details;
Application Tier
CPU: Two servers each with two dual core Intel Xeon 3 MHz processors
OS: Red Hat Enterprise Linux AS release 4 update 4
Application Server: WebLogic Server 9.2
JVM: BEA JRockit(R) 5.0 JDK (R27.1.0-63)
Database Tier
CPU: One server with four dual core Intel Xeon 3.4MHz processors
Database Software: IBM DB2 Universal Database v8.2.6
OS: Red Hat Enterprise Linux AS release 4 update 4
This system produced 1,466.29 SPECjAppServer2004 JOPS@Standard
This is the first submission using dual-core Woodcrest processors, which helped give it the highest per-core result to date.
Despite the relatively small hardware configuration it is the 10th highest throughput we've seen, which is impressive given some earlier scores that needed significantly more hardware to achieve a similar result. For example, in November of 2004 IBM submitted a configuration that needed 20 app server CPUs (Xeon/SuSE) to process 1,343.47 JOPS. That's five times as many app server CPUs yielding less throughput when running the same workload. Again, those are old results using different hardware and software, but it is interesting to see how far these products have come in two years.
WebLogic Server now holds the top two spots on an absolute basis, the highest throughput for 2 and 4 CPU configurations and the highest JOPS/core.
All of the results can be see here.
Past updates of benchmark results:
Analysis of SPECjAppServer2004 Results as of September 2005
Analysis of SPECjAppServer2004 Results as of January 2006
Analysis of SPECjAppServer2004 Results as of February 2006
Analysis of Application Server Benchmark Results as of March 2006
SPECjAppServer is a trademark of the Standard Performance Evaluation Corp. (SPEC). Competitive numbers shown reflect results published on www.spec.org as of November 30th, 2006.
First, let me say that I'm not harping on Oracle to spread FUD about a competitor. The growth numbers they publish give the impression that there is a vast install base hardening their products, creating a large market opportunity for ISVs to sell into. I simply want to help end users and ISVs see some of the tactics in play that may affect their purchase and porting decisions.
There are two factors that I suggest tracking. First, it's important to understand how they get to these numbers. Second, I would strongly suggest not assuming that Fusion Middleware numbers = application servers and related products (in the BEA context).
Oracle has all guns blazing right now. If you listened to their Q1 results you would have heard that their middleware business grew more than 50%... "faster than BEA". This has been a reoccurring theme for them for the last few quarters.
The way Oracle gets to this number is by adding in some of the acquired products from Siebel, PeopleSoft, Collaxa, Oblix and others into their middleware revenues, making their year over year comparisons look huge.
Skeptical? Here are a few data points to consider.
Oracle just announced that going forward they will not break out the Fusion Middleware growth number any more. All we will get is a combined Database and Middleware number. I would suggest that this is because they will anniversary the acquisitions, making their year over year growth numbers come down to something far less marketable.
Some Wall Street analysts are starting to untangle and publish estimates of Oracle's middleware growth rates and revenues that are far more modest... as in low double digits on a relatively small base. Denny Fish from JMP Securities and Charlie Di Bona at Sanford Bernstein have recently done some good analysis on this.
If you are in the market for an application server and simply want to know who the leaders are I would suggest looking at the Gartner Market Share data for J2EE app servers in 2004 and 2005 (it's in the 2005 Application Integration and Middleware Market Share report from June, 2006).
If you want to see application server performance benchmarks, see the SPECjAppServer2004 results, although Oracle and JBoss haven't published results for their app servers for some reason ;-)
If you want to see which application server developers prefer, see the new Evans Data Corporation survey titled "Professionals Choice: Application Servers 2006 User Ranking." This report surveyed more than 700 developers on a variety of features and characteristics of application servers.
There is do doubt that Oracle is a large, formidable competitor. That said, their bark is often bigger than their bite.
Oracle recently issued a press release, claiming to be "growing faster than BEA", with middleware licenses increasing 24% last quarter, compared to BEA's 18%. Before taking this claim at face value, consider the following.
1. The way Oracle breaks out their revenue numbers makes it impossible to confirm or deny any of their middleware growth claims. Oracle only breaks out two lines of software licenses 1) Applications and 2) Database & Middleware. Revenues can get shuffled around on a fairly arbitrary basis. For example, when a customer signs an ELA, how do they assign revenues to each product? Some wonder if a disproportionate amount is allocated to where ever they want to show growth. Oracle used to criticize IBM for playing this game.
2. Their reported middleware growth rate includes an unknown amount of legacy and other products that BEA does not compete for. To make a reasonable comparison to BEA you would need to remove Oracle Forms, their Business Intelligence products (i.e. Siebel Analytics + Oracle Business Intelligence Suite), their security products (Oblix), data hubs and other miscellaneous products from Siebel, Peoplesoft and other acquisitions that are in product spaces that BEA does not compete for. This in itself should reduce their 24% growth rate well below BEA's.
3. Market share is an aggregate of install base and new customers. BEA has been amassing significantly more middleware customers than Oracle over the last ten years. Even if you doubled Oracle's middleware revenues it would take them years to catch up to the size of BEA's install base/market share.
Wall Street, who tracks these numbers closely, doesn't seem to buy their story either.
At the end of the day Oracle is blurring the lines to try to make it look like their current application server, integration, portal and SOA products are doing better than they really are.
The battle for the SOA platform is well underway. Today IBM announced a variety of new SOA products and services. For those of you keeping score, here are a few points to keep in mind.
The Performance Benchmarks Tell an Interesting Story
Performance and scale are core to any infrastructure platform, especially when using XML and Web services.
At this point it looks like IBM has conceded the SPECjAppServer2004 benchmarks to BEA. BEA currently owns the top three throughput results and has shown that WebLogic can process the same number of transactions on half the hardware as WebSphere. IBM hasn't published a benchmark since October 2005.
A summary of the performance benchmark results can be seen here.
The Product Reviews Tell an Interesting Story
Network Computing did a detailed Enterprise Service Bus product review in March 2006. The ESB is generally considered to be core to SOA. IBM's ESB came in 6th out of 8 products reviewed. BEA's AquaLogic Service Bus, which was launched in July of 2005, took 1st place. The full details can be seen here.
Network Computing also did a BPM product review in July of 2005. Fuego, which is now AquaLogic BPM, took first place. IBM did not compete. The details can be seen here.
IBM's Q4 Financial Results Tell an Interesting Story
IBM published their Q4 results on January 17th. IBM Software grew 0% in Q4. The WebSphere family grew 4%. Lotus grew 2%. Tivoli grew 3%. Rational was negative 2%.
This chart is from their earnings call, available on ibm.com.
These numbers include license and maintenance revenues. While it's not possible to get an exact license number for each product family, it is important to realize that maintenance represents a significant portion of any revenue number because it is an aggregate of new and install base customers paying the annual fee. For this reason it is safe to assume that WebSphere and other key branded license revenues are shrinking, not "gaining share" as they say.
BEA reported Q4 earnings on February 23rd.
*For the fiscal year, revenue was a record $1.2 billion, up 11% from last year.
*For the quarter, total revenue was a record $341 million.
*For the quarter, license revenues were up 18% year/year. To compare this number to the IBM numbers you need to add maintenance revenues back in, which makes the growth rate significantly higher.
*BEA had a record of 31 deals over $1M.
*Cash Flow from operations was over $100M for the first time.
A big reason for this financial performance is because of AquaLogic, BEA's platform for SOA. In Q4 BEA reported that AquaLogic represented 10% of license revenues, which surprised many people for a brand launched in June of 2005.
The product family is made up of internally developed products such as the Service Bus, the Data Services Platform and Enterprise Security as well as the recently acquired products from Plumtree and Fuego.
For a longer range view of BEA's financial performance, see this blog entry.
The Stock Chart Tells an Interesting Story
Wall Street is also voting with their dollars. The chart below is a trailing 12 month view of BEA and IBM stock prices, which reflect the perception of performance and potential of each company.
On March 21st BEA hosted the sixth annual Financial Analyst Day featuring presentations by Chairman, Founder and CEO Alfred Chuang, Chief Marketing Officer Marge Breya, President of World Wide Field Operations Tom Ashburn, Senior Vice President of Americas Sales Rich Geraffo and Chief Financial Officer Mark Dentinger. The day covered BEA's product and go-to-market strategies as well as customer presentations by Verizon Wireless, Visage Mobile and Pratt & Whitney.
The audio and slides are available at bea.com/investor. We have also posted the AquaLogic, WebLogic Platform and WebLogic Communications Platform presentations that were delivered the following day.
I realize that this isn't typical dev2dev technical content, but figure that some of you are interested in BEA's business strategy.
There were two new SPECjAppServer2004 publications since my last recap of the results. These were mid range configurations run by HP.
The first used WebLogic Server 9.1, JRockit 5, RedHat Enterprise Linux 4 Update 2 running a single application server with four Itanium2 processors. The database was Oracle Database 10g Enterprise Edition Release 10.1.0.4 running on HP-UX 11i with four Itanium2 processors. This configuration processed 542.17 SPECjAppServer2004 JOPS@Standard. The full configuration details can be seen here.
The second configuration used BEA WebLogic Server 9.0, HP's VM, HP-UX rx4640 running a single application server with four Itanium2 processors. The database was Oracle Database 10g Enterprise Edition Release 10.1.0.4 running on HP-UX 11i with four Itanium2 processors. This configuration processed 538.03 SPECjAppServer2004 JOPS@Standard. The full configuration details can be seen here.
HP ran a similar configuration in September of 2005, using WebLogic Server 9, HP-UX, Oracle and Itanium2 processors but only got 471.28 SPECjAppServer2004 JOPS@Standard. The new results are about 14% better.
We now have a total of 19 configurations to compare. Ten are running WebLogic Server, including the top three scores, which are published on a diverse set of hardware, OS and VMs.
Five use WebSphere but IBM hasn't published a score since October, 2005. Four use Sun Java System Application Server. Still no results from Oracle or JBoss, almost two years after the first SPECjAppSever2004 score was published.
All of the results can be see here.
SPECjAppServer is a trademark of the Standard Performance Evaluation Corp. (SPEC). Competitive numbers shown reflect results published on www.spec.org as of March 13th, 2006.
Network Computing just published a technical primer and lab based product review of ESBs, seen here.
They do a nice job of explaining the role of the ESB and their test methodology as they look at products from BEA Systems, Cape Clear Software, Fiorano Software, IBM, Oracle, Software AG, Sonic Software and TIBCO Software.
If you want to cut to the chase, you can skip ahead to the final Report Card.
I though this article about an RFID project at McKesson was really interesting.
RFID will create some really interesting use cases over the next few years. We see it as another infrastructure play as companies, governments and ISVs build out RFID enabled applications.
In 2005 BEA purchased ConnecTerra to fill out our offering and launch a product in this space. It will be an interesting area to watch.
BEA, IBM and others have been submitting projects to open source over the last several years. Oracle is going the other way, trying to bring them in house by acquiring JBoss, Sleepycat and others. Smart move or colossal mistake?
Oracle
Oracle has a spotty record in the middleware space. In the 1999 timeframe they had Oracle Application Server (OAS). This was a previous generation app server based on Oracle Forms. A few years later they replaced it with the first version of Oracle 9iAS, another internally developed application server with more support for Java and J2EE. Oracle claimed that this product was wildly successful, but eventually discontinued it when they rolled out a new version of 9iAS based on the Orion application server (which they licensed from Iron Flare). They forked the code base and evolved it into their current 10g AS offering.
Now Oracle is rumored to be buying JBoss, their fourth application server. Will they use JBoss for a low end offering or replace 10g AS entirely? It's hard to say at this point.
Having a J2EE implementation is easy. Creating a bullet proof middleware platform is a different story. Oracle has shown that they are not capable of developing or bringing forward acquired middleware products.
Oracle talks about standards leadership but took until January 2006 to ship a J2EE 1.4 production supported version of their app server, two and a half years after the spec was completed. (They've been marketing J2EE 1.4 support for a long time but you needed to read the fine print to see that it was only a "developer preview".) They have also been touting their EJB 3.0 implementation but the developer reviews have been anything but stellar.
As Oracle knows in the database space, performance and scale differentiate enterprise software products. SPECjAppServer2004 is the industry standard benchmark for comparing application server performance and scalability. Neither Oracle nor JBoss have published any performance benchmarks. The current results are summarized here. On the tooling front, JDeveloper is not based on Eclipse, the de facto Java IDE.
Oracle also has credibility problems in this space. They claim to be #2 in middleware market share but have no credible data to back it up. The industry analysts certainly don't agree with this claim. In the packaged application market, Oracle has gotten themselves into hot water several times lately by misstating market share data from both Gartner and Stratascope.
The information they give about competitive wins has been shown to be untrue, as have their performance and TCO claims.
Marketing aside, real deals come down to real products and projects. The email below is what I'm seeing from within BEA. This was a Portal deal.
-----------------
(I removed the customer and rep names because I don't have their permission to publish them.)
1/30/06
BEA Team - Wanted to update you that we have received fantastic news from XXX - we have won the Technical Evaluation against Oracle!! Over the next several weeks, we will be continuing the battle as this moves beyond the technical discussions. This showcases not only the strength of our product but also the strength of our team. Some details we've learned of the comparison of the BEA and Oracle POCs...
*Oracle took 5 days to install while John took 30 minutes
*Installation for Oracle took attempts on three different machines because they needed to change registry properties while John just un-tar-ed the install on one machine
*John's POC development was accomplished in a little over a week, as the first week was spent fighting with firewall and connectivity issues.
*John also spent time doing knowledge transfer to XXX
*Oracle's was only given a week for their POC (since John had worked through the connectivity issues). However... it took the full two weeks by a team of 5 developers onsite (and who knows how many offsite).
*Oracle had no time for knowledge transfer and...
*NONE of the functionality outside of the Oracle connections were implemented!!!! No integration to Stellent, no integration to Clarify. Nothing outside of iSupport and Contracts.
---------------
JBoss
BEA looked at JBoss as it was shopped around this winter. On one hand JBoss did a good job of building industry buzz around their product and the concept of "Professional Open Source". On the other hand, like Oracle, the leaders of JBoss greatly exaggerated the effect they were having on BEA's core business.
Customers liked the JBoss concept because it promised to lower TCO by not charging a license fee. They also liked the fact that if they stuck to the J2EE APIs, they could migrate to a commercial product like WebLogic when they needed to. These two factors drove a lot of interest in the product.
In reality JBoss is a very basic application server. As you can see here, JBoss does not scale very well. It is also light in security, administration, high availability and productivity features. Each of these has profound TCO implications that overshadow the cost of a WebLogic license. I blogged about the relative cost of application servers and did a crude TCO analysis here.
Many customers learned the hard way that the parachute of portability is not always there. With the best of intentions they have used JBoss and tried to stick to the J2EE APIs but ended up using proprietary features like the portal, O/R mapping, caching and BPM, forgoing the ability to port the application to another application server.
Finally, the "openness" of the JBoss community has been questionable from the start. JBoss is a for-profit company with licenses, trademarks and other assets that they have aggressively enforced. This is fine for most commercial companies but goes against many of the open source community values.
Oracle + JBoss
Can Oracle keep the JBoss community together or will all of those developers and users join the Apache Geronimo effort?
I know one thing for sure; the JBoss community is not going to allow their efforts to be used as a low end seeding vehicle, solely for up selling Oracle products. This has been a religious battle for them over the years. It's going to be a fascinating culture clash. I see the JBoss community disappearing in a poof of smoke the second Oracle, or any commercial vendor, tries to monetize it. We just saw this happen with SuSE and Novell.
Many open source developers are going to look at Oracle's track record in the open source database market. A few months ago Oracle bought InnoDB to get the storage engine embedded in MySQL. MySQL then started working with SleepyCat to replace this technology. Now Oracle is rumored to be buying SleepyCat. It looks like Oracle is not embracing open source as Ellison says, but instead buying it to kill it.
Community aside, the combined product portfolios almost completely overlap.
*Which J2EE container will they bring forward, Oracle 10g or JBoss App Server?
*Which one will Oracle applications run on?
*When will they publish performance benchmarks for either?
*Which O/R Mapping technology will they bring forward, TopLink or Hibernate?
*Which BPM products will they bring forward, Oracle BPEL Process Manager or JBoss jBPM?
*Which Portal will they bring forward, Oracle Portal or JBoss Portal?
*Should customers use JBoss Transactions (from Arjuna)?
*What are the licensing implications for Oracle ISVs who want to use JBoss products?
*What are the licensing implications for JBoss developers and ISVs when Oracle brings them into the fold?
Product Fusion was ambitious enough. Acquiring these products and customers will decrease their odds of ever getting to a single, unified platform.
BEA's Open Source Strategy
BEA is taking a different approach to open source. We call it our "blended" strategy which brings together the best of commercial and open source products into a single, rationalized, tested and supported platform. On the tooling and framework front BEA is supporting the Eclipse IDE as well as Spring, Struts and Beehive.
If a customer is using a non-BEA application server, we will sell and support products that run on or around it. The new version of Workshop deploys to JBoss, Tomcat, Jetty, WebSphere and Resin, as do some of the AquaLogic, RFID and other BEA products.
The net/net
As this plays out we will see many marketing claims and accusations fly. The undeniable facts are the results published by the companies involved.
SAP has seen great results in their applications business after the Oracle acquisitions of Peoplesoft, JD Edwards, Retek, Siebel, etc. They recently announced 18% license growth for 2005.
BEA has never been on stronger financial footing. See Gartner's "A Strong Quarter Brings More Confidence in BEA Systems" which was published after our Q3 Earnings. Since then BEA has pre-announced our Q4 results, which will be fully disclosed on February 23rd.
Will this acquisition work out? Only time will tell but I'm betting on BEA to continue to be the focused leader in this space.
We have only see one new submission since my last recap of the SPECjAppServer2004 results. The latest result was submitted by Sun, who is the first to break the 4,000 JOPS@Standard barrier.
Running WebLogic Server 9.0 on 56 cores on 7 chips they accomplished 4098.77 JOPS@Standard. All of the configuration details can be seen here.
We now have seventeen configurations and scores to compare.
WebLogic Server is used in eight configurations, including the top three scores.
Five configurations use WebSphere, with a high score of 2921.48 JOPS@Standard.
Four configurations use Sun Java System Application Server, with a high score of 1201.44 JOPS@Standard.
Oracle and JBoss have still not submitted results.
All of the results can be see here.
SPECjAppServer is a trademark of the Standard Performance Evaluation Corp. (SPEC). Competitive numbers shown reflect results published on www.spec.org as of February 7th.
Oracle and SAP say that enterprise software is boiling down to three or four vendors. Why bother with all of the moving parts, they ask, while citing industry statistics that make it sound inevitable. I'm placing my bet on the other end of the spectrum. The next few years will see the rise and fall of technology empires and will be the most exciting years the industry has ever seen.
The bubble years of 1999 through 2001 focused on start ups and IPOs. The bubble bust years of 2002 through 2005 pushed the pendulum to the other end of the spectrum, favoring the software mega-caps like Microsoft, IBM, Oracle and SAP. In 2006 we see the pendulum coming back to the center, with vendors like Apple, Adobe, BEA, Business Objects, Red Hat, Salesforce and others benefiting from focus, not breadth. This ebb and flow will always be the case in the technology industry.
While tech investment, spending and stock prices have been lackluster for the last few years, the train of innovation continues. This will carry on the decades long tradition of technology disruption and upheaval that will inevitably replace today's 600 pound Gorillas with a new generation of leaders.
For example, the way we interact with applications will continue to evolve in 2006. While Web applications have been the story of the last decade, the interaction model was a huge step backwards from the rich client. We take it for granted, but if you think about it, a browser displaying HTML makes a lousy user interface.
2006 is going to see some real progress in the march towards rich Internet applications; the best of client/server and Web applications combined. New standards and frameworks like Java Service Faces, AJAX, Avalon and portal products will clash, combine and give birth to the next generation user interface. Adobe's acquisition of Macromedia and what they do with Flash, Flex and PDF is another one to watch.
Just when you think it's exclusively a Windows world, Macintosh comes back with a vengeance. While coffee shops, schools and design shops have been their traditional domain, I'm now seeing them toted around en mass. A CIO of a large company recently told me that she is considering a full swap out of Windows desktops for Macintoshes. Combining this with the recent partnership with Intel and the iPod phenomena that places Apple squarely in the eye of the digital entertainment tornado makes it clear that they have successfully redefined the rules of the desktop computing market.
Beyond the age old Windows vs. Macintosh debate, there are rumblings of Google rolling out a Windows-less PC and hosted desktop productivity applications that will compete with Microsoft Office. Blackberrys, Nokia's smart phones and others will also continue on their march to mass market computing platforms.
All of this has huge implications for Microsoft. Without a homogenous, Windows based desktop, companies are not going to use the proprietary rich client features in Vista or let Microsoft control the standards that browsers aggregate and render. This will eventually wind down the Windows monopoly.
In the packaged application market, 2005 was the 'house cleaning' year for the client/server products that were brought to us throughout the 1990's to get ahead of the Y2K problem.
It all started a few years ago, when Oracle found themselves in the slow growth database market with increased competition from open source and Microsoft at the low end and IBM DB2 on the high end. Investors raged and Oracle went into action, promising 20% earnings growth for the next five years. This set off the consolidation of Peoplesoft, JD Edwards, Siebel, Retek and others to kick start their failed attempts to build and sell the applications themselves.
Entering 2006 SAP is still king of the apps and appears to be distancing their lead. Their applications also drive more database licenses than anyone other than Oracle themselves. This may change if SAP has anything to say about it. After scuttling SAP DB to MySQL, they are now paling up with Microsoft, who has a bulls-eye on Oracle's database business. SAP is endorsing Microsoft's latest version of SQL Server and is collaborating with them on a product that ties Microsoft desktop applications to SAP back office applications. This will be interesting to watch.
Middleware continues to grow in significance, particularly as we see the SOA go mainstream. Everyone agrees that a constellation of applications will be available over the Internet, exposed as Web services, spanning everything from ERP to payroll, banking and personal finance applications connecting to other back end systems or your desktop, iPod and phone. Applications that use RFID sensor networks, location based services and voice, video and data over IP will usher in countless new use cases that we aren't even dreaming about today. The technology is here, ripe for creativity and new business models to disrupt the status quo.
Few people have picked up on the threat that this represents to the big packaged application and platform vendors who are used to selling broad, proprietary suites. First, the whole point of SOA is to facilitate interoperability of heterogeneous systems, regardless of what hardware, operating system, database or programming language the service was built on. This will make it harder for Oracle, SAP, Microsoft or IBM to become a single-vendor standard within a customers' IT environment.
Second, the application vendors are being forced to make their applications accessible as standards-based Web services (open APIs and protocols), which will loosen the grip they traditionally had on their customers; forcing them to only use their add-on products once they are locked in.
Third, a new generation of hosted services, like CRM functionality from Salesforce.com, commerce functionality from Amazon, Yahoo and eBay, Web analytics functionality from Google and others are being made available programmatically, not just through a browser. This allows developers to integrate remote functionality into their applications, which reduces their dependency on traditional packaged applications and all of the headaches that come along with them. The combination of these factors will decentralize the application market, not centralize it the way Oracle is trying to do.
In the middleware/SOA space, BEA is the focused technology leader. IBM has the broadest product and services portfolio. In 2006 JBoss, who has been the low cost option, will see real competition from the Apache Geronimo project, which will offer a similar product from a non-profit under better licensing terms. Oracle and SAP will claw it out for the number five and six positions with Fusion and Netweaver. Their success will be limited to the niche segment of customers who buy into a single vendor strategy, not the vastly complex and heterogeneous verticals like telecommunications, financial services and government. Sun, who has been trying to gain credibility in the middleware market for years, has thrown a hail-Mary by open sourcing and giving away their middleware products as the final chapter in the NetDynamics, Forte, Netscape, iPlanet, SunOne, SeeBeyond, Java Enterprise System saga.
Beyond the brand names, SOA will create a slew of new problems that will spin up a cycle of investment, start up companies, rapid innovation and new products that will either fail, be acquired or establish new players. Examples include Amberpoint, Fuego and Systinet (who was just bought by Mercury).
The programming languages, models and frameworks also continue to evolve. The Java community has responded to criticism of complexity with Java Enterprise Edition 5.0 and a slew of application frameworks, all focused on simplifying the development of enterprise applications. Watch for EJB 3.0 and JSF implementations in 2006, as well as continued adoption of Spring, and other application frameworks. Scripting languages like PHP, Python, Perl and especially Ruby are also stirring up interest. Windows Vista, with its new version of the .Net framework, is supposed to ship in 2006, although after two years of delay no one is quite sure what effect it will have.
On the tooling front, Java developers have coalesced under the Eclipse umbrella. The door is still cracked open for NetBeans and IntelliJ and is closing on Borland. The pluggable Eclipse IDE will bring order to the otherwise disjoined tools space and create opportunities for more vendors to play.
We are also seeing movement on the hardware side. For the last few years we have seen x86 running Linux and Windows disrupt the traditional RISC/UNIX server market, reducing cost of acquisition but expanding data center and management costs. 2006 will see a move to increase server utilization and reduce management costs by employing virtualization technologies and multi-core systems to reduce datacenter floor space, heat dissipation and power consumption.
Intel is pushing Itanium and dual-core Xeon, AMD is pushing dual-core Opteron and Sun launched their 8 core SPARC processors. Smaller focused players like Azul Systems, Egenera and others are also making headway. On the software side of virtualization, VMware is red hot but has their work cut out for them with Intel, AMD, IBM, HP, Sun, Rehat and Suse aggressively backing Xen, an open source virtualization project that is scheduled to be enterprise ready in 2006.
And this is just the tip of the iceberg. Security, compliance/risk management, disaster recovery, storage, business intelligence, outsourcing and other areas will also change the way we think about enterprise computing.
The best part is that this isn't technology for technology sake. Look at what RFID will do for supply chains, retail and manufacturing, what VoIP will do to the telcos, what streaming video and music will do to the content creation and distribution industries. Anyone who thinks a single vendor will own the underlying machinery that makes these applications possible is in for a surprise.
Things have heated up in the SPECjAppServer2004 benchmark wars since my recap of the results in September. Since then we have seen some monster configurations spun up, each trumping the last world record. All of the current results can be seen here.
To recap the last few months, IBM published a new SPECjAppServer2004 World Record using WebSphere 6 in October, processing 2,921.48 JOPS@Standard. All of the configuration details can be seen here.
In November Sun trumped IBM by using WebLogic Server 9.0, processing 3,328.80 JOPS@Standard. All of the configuration details can be seen here.
In December HP struck back, also using WebLogic Server 9.0, breaking the World Record again, processing 3,734.68 JOPS@Standard. All of the configuration details can be seen here.
As of January 5th, 2006 we have 16 SPECjAppServer2004 results to compare. Seven of them use WebLogic Server 9.0, five use WebSphere 5/6 and four use Sun Java System Application Server. Oracle and JBoss have been conspicuously absent from the competition. Neither has published a benchmark result yet. Oracle has been using old benchmark data to justify their performance and TCO marketing claims, which I blogged about here.
SPECjAppServer is a trademark of the Standard Performance Evaluation Corp. (SPEC). Competitive numbers shown reflect results published on www.spec.org as of January 5th.
SPECjAppServer2004, the recently updated version of the industry standard benchmark, is far more comprehensive than SPECjAppServer2002, putting the app server under a rigorous workload that is much closer to a real world enterprise application than the previous version. The new workload tests the EJB container, including message driven beans and CMP, the Web container including Servlets and JSPs, the message bus, the transaction manager, database connectivity and other subsystems of an application server.
For background on SPECjAppServer, I would highly recommend this entry by Samuel Kounev- SPECjAppServer2004: The New Way to Evaluate J2EE Performance.
Analysis of SPECjAppServer2004 Results
So far results have been published by BEA, IBM and Sun. Oracle was active in SPECjAppServer2002, so one can only infer why they haven't published a SPECjAppServer2004 result yet. JBoss is also part of SPEC but has not published a score yet.
The new result from BEA sets a new scalability world record, beating the previous record which was already held by BEA.
This result also demonstrates the linear scalability of WebLogic Server 9.0. The previous configuration had WebLogic Server 9.0 and JRockit 5.0 running on five HP servers, each running RedHat Enterprise Linux 4 Update 1 on 2 Intel Xeon processors (a total of 10 CPUs). The resulting throughput was 1,374.11 JOPS. If you divide the throughput by the number of app server CPUs you get 137 transactions per second per CPU (1,374.11/10).
In this configuration the performance engineering team added a 6th server to the application server tier, bringing the total number of CPUs up from 10 to 12. They also added four CPUs to the database server. If you compute the number of transactions per app server CPU again you get 138 transactions per second per CPU (1,664.36/12), which shows perfectly linear scalability.
Competitively speaking, these numbers show that WebLogic Server 9.0 is processing twice the transactions per CPU as WebSphere 5.1, which has a best score of 1,306.44 on 22 application server CPUs. This is a key take away. Even if WebSphere costs less per CPU that WebLogic, you need more than twice as many licenses to achieve the same throughput under the same workload.
WebSphere 6.0 does not appear to have a huge performance improvement. The highest throughput configuration based on WebSphere 6.0 achieved 1,343.47 transactions using 20 application server CPUs.
Finally, the Sun numbers are a perfect example of slower software consuming massive amounts of hardware and software licenses. The one scaled up Sun Java Application Server 8.1 result is processing 1,201.44 transactions but it took 26 application server CPUs to get there. This is a perfect example of the potential for free software, such as Sun's app server or JBoss, to drive costs up by needing significantly more hardware.
Other Scalability Benchmarks
If industry standard benchmarks aren't your cup of tea, see the whitepaper that Epiphany published, showing their application running on WebLogic Server 8.1 on HP Integrity Servers, scaled up to 10,000 users doing 1,152,000 transactions per hour with a response time of .53 seconds. It is from December 2004, so it is a little dated, but still makes for an interesting read.
For all of the details on all of the SPECjAppServer2004 results click here.
This entry lays out the big picture of BEA's financial performance since the company went public in 1998. It shows the strength in our operating model, how we have weathered a difficult period for all enterprise software vendors and have managed to increase funding for R&D every year to stay ahead of our competition.
The chart below shows BEA's annual revenue since IPO. As you can see, BEA established a base with Tuxedo, grew rapidly with the adoption of the Internet and has sustained revenues at those levels through the industry wide downturn of the last few years. There was only one year when revenue slightly declined, in FY 03. BEA returned to growth and passed $1b in revenue in FY 04. In FY 05 total revenue grew 7% to $1.08 billion.
Revenues
[BEA's fiscal year ends January 31, so the FY designations are roughly one year off from the calendar year - FY 05 is the last 11 months of 04, and the first month of 05.]
The white bar at the bottom of the chart is license revenue. The red bar at the top is services revenue, which is a combination of consulting and customer support.
License revenue grew by about 1% in FY 04, then declined by 7% last year. Our number one priority has been to get license revenue back to healthy growth levels, which the new product and go to market strategies have already turned the corner on this year.
In the earlier years, customer support represented about 25% of the red bar, but has grown steadily over the years. In the later years customer support revenues make up about 75% of the red bar. Customer support revenues grew 24% last year, to $463 million. We have had very stable, high renewal rates on customer support along with stable pricing. This tells us that customers are having success deploying projects on our software, that they value the software updates included in our support contract and that we don't have shelfware. Also, this improves the stability and predictability of our total revenue, which in turn improves the stability and predictability of our profitability and cash flow.
BEA has reported results for the first 3 quarters of our fiscal year 2006. We are off to a good start. Total revenue for the third quarter was our highest ever, $292 million, up 10% from last year. For the first 3 quarters of the year, total revenue was $858 million, up 9% from the first 3 quarters of last year.
More importantly, in Q2 we achieved our goal of returning to year/year license revenue growth. In Q3 we accelerated the growth to about 6%. These results show that the core business is strong and how the new products are starting to contribute in a meaningful way.
Non-GAAP Operating Profit
Non-GAAP operating profit follows a very similar pattern to revenue - rapid growth in 1999-2001 and stability since the industry downturn. As with revenue, there was only one year that profit declined (2003) but has rebounded since. Profit grew 12% last year, to $220 million.
We are profitable on a GAAP-basis as well, with GAAP operating profits of $175 million in FY 04 and $195 million in FY 05. We think pro forma is a better measure - with one-time restructuring charges in FY 02, the GAAP operating profit shows a huge improvement in profitability that could be misleading. The pro forma picture is more closely aligned with what is happening to the long-term model.
In the first three quarters of fiscal 2006 we had record profits by both GAAP and non-GAAP measures. GAAP operating profit was $148m, up 9% and non-GAAP operating profit was $161m, up 3%, compared to the first 3 quarters of last year.
Operating Cash Flow
Our operating cash flow has also been very strong - growing rapidly in the 1999-2001 time frame, then sustaining around the $200 million per year level through the industry downturn. Last year cash flow grew 25%, from $213 million to $267 million. We have generated $183 million in operating cash flow through the first three quarters of this year.
Cash Balance
Our stable revenue and healthy operating model have led to a steady increase in cash balance. At the end of our last fiscal year we had $1.6 billion in cash on our balance sheet. At the end of the third quarter our cash position was down to $1.4 billion due to closing five acquisitions this year, including Plumtree. In addition, BEA repurchased $161 million of stock in FY 05 and repurchased $180 million of our stock through the first 3 quarters of this year.
BEA has $550 million of publicly-traded convertible debt. It is due in December 2006. Although the conversion price of about $35 per share is well above our current stock price, our cash balance and cash flow give investors confidence in our ability to pay the debt, which trades very close to face value.
BEA also has about $215 million in unsecured long-term debt outstanding, due in October 2008.
So, in evaluating our cash position, it is reasonable to look at the balance as either $1.4 billion in gross cash, or $680 million in cash net of debt.
Our cash balance and operating model put us in a strong position to invest aggressively in R&D.
R&D Investment
BEA has increased the R&D investment from $90 million in fiscal 01 (calendar 00) to $120 million the next year - a 33% increase in R&D investment in the face of the industry wide recession. We have continued to increase our R&D investment since that time, with a 60% increase in R&D investment from FY 01 to 05.
In the first 3 quarters of this year we have re-accelerated our R&D investment. R&D is up 20% compared to the first 3 quarters last year.
This R&D investment is important. We know that customers expect us to be innovative - not only to stay ahead of the curve, but to define the curve. We expect the same of ourselves. Our cash position and operating model have given us the ability to invest aggressively at a time when other software companies were cutting R&D investment in the recession.
| Sun | Mon | Tue | Wed | Thu | Fri | Sat |
|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | ||
| 6 | 7 | 8 | 9 | 10 | 11 | 12 |
| 13 | 14 | 15 | 16 | 17 | 18 | 19 |
| 20 | 21 | 22 | 23 | 24 | 25 | 26 |
| 27 | 28 | 29 | 30 | 31 |
January 2007
November 2006
September 2006
April 2006
March 2006
February 2006
January 2006
December 2005
November 2005
October 2005
September 2005
August 2005
July 2005
New Application Server Scalability Benchmark Record
Analysis of SPECjAppServer2004 Results as of November 2006
Dissecting Oracle's Middleware Growth Claims... again.
Six Cost Reduction Strategies for 2004
This article is focused on several strategies for reducing cost and risk in 2004. They are based on real-world industry trends around the technologies and skills available in today's market. Mar. 10, 2004
