DBMS Interview - October 1995
"It's not just a database company anymore." So says Sybase's Dennis McEvoy, as he pulls out a slide showing that Sybase Inc. ranks sixth among the world's largest software companies, with $826 million in 1994 revenue. With entries in the DBMS, middleware, and tools categories, Sybase now wields the combined technical and marketing capabilities of the diverse companies it has acquired and integrated. Those companies include Gain Technology Inc. (multimedia), Micro Decisionware Inc. (database connectivity), Powersoft Corp. (application development), Expressway Technologies Corp. (query acceleration), Watcom International Corp. (languages and DBMSs), Complex Architectures Inc. (message-oriented middleware), and SDP Technologies Inc. (CASE/modeling). Wrap those technologies around a solid and scalable DBMS platform and you've got a formidable client/server solutions portfolio.
It's the "scalable" part of the equation that the new Sybase SQL Server 11 will set out to solve. In the past year, concerns about SQL Server 10's quality and its multiprocessor scalability helped flatten the company's revenue and stock value. Major ISVs such as Peoplesoft Inc. (Pleasanton, Calif.) and SAP AG (Walldorf, Germany) declined to port their applications to SQL Server 10, while announcing support for Microsoft SQL Server 6.0. For Sybase, there could be no sharper rebuke.
Now, Sybase has hinged its future on the performance and scalability of SQL Server 11. McEvoy, who serves as the company's products group vice president, states clearly and confidently that Sybase will succeed. After 25 years in the computer industry, McEvoy knows a defining moment when he sees one.
McEvoy joined Sybase in 1994 as vice president of tools technology. After Sybase acquired Powersoft in October, he became vice president of the products group, where he manages the development of Sybase server and connectivity products, including engineering, product management, product marketing, quality assurance, and technical publications. Before joining Sybase, McEvoy ran Cooperative Solutions, which he cofounded in 1989. That company created Ellipse, one of the earliest client/server environment for mission-critical OLTP systems. That company and product were later sold to Bachman Information Systems. Before starting Cooperative Solutions, McEvoy served as vice president of the software division at Tandem Computers, where he spent 14 years. Before that, he worked at HP, where he collaborated on the original MPE operating system. When McEvoy joined HP in 1969, the company had only 17 people working in software.
DBMS Editor in Chief David Kalman and Features Editor Theresa Rigney recently interviewed McEvoy at Sybase's Emeryville headquarters. An edited transcript of the interview follows.
DBMS: What is the Sybase product strategy?
McEvoy: We're focused on the use of technology in three primary market segments: OLTP and mixed workload for running your business (mixed workload means OLTP and real-time decision support); data warehousing and decision support for predicting your business; and "mass deployment" for extending your business to the endpoints of the organization.
We have a single architecture comprising database, middleware, and tools, but we don't have a "one size fits all." Each of these market segments has different requirements, so they require products that are tailored for each of them. We're developing tailored products that are consistent across the architecture and consistent in the APIs, but are specifically optimized for these three broad segments.
Today, a majority of revenue comes from OLTP. Data warehousing is clearly an explosive market, and 1995 is the first year that showed any kind of substantial revenue within that market. Mass deployment is a market in which Sybase believes the revenue opportunity will build over the next 36 months.
In each area, we have products for database, middleware, and tools. These include the Powersoft tools family, which is a de facto standard. In the middleware category, we have the Enterprise Connect family, which is based on the MDI technology. The MDI name is going away. We have two brand names now, a Sybase brand for database and middleware and a Powersoft brand for tools.
In the database category, we have the SQL Server database family, focused on price/performance and scalability, however they're defined in each of these segments. For example, in the mass deployment segment, scalability refers to the size of the footprint, as opposed to how many processors you can support. In data warehousing, scalability involves how quickly planned and unplanned queries complete, how big a database you can support, and how many users you can support. In OLTP, the concern is price/performance and scalability.
Clearly, the customers' needs in each of these areas are different. For example, in mass deployment, the issue is whether you can run a DBMS without a DBA. Can I put SQL out on a mobile computer used by a salesperson who is not supported by a systems administrator? Can I do it on the same Windows machine that I was going to buy, instead of having to buy a 16MB or 32MB machine (RAM) to support an RDBMS?
For example, Personal Oracle requires 24MB of memory. Oracle says 16MB, but you can't run it on that. SQL Anywhere [the new name for single-user Watcom] requires only 1MB of memory, and 4MB on the disk. Contrast that with the one-size-fits-all approach of trying to have the same code line run on 128-processor MPP machines and also fit on Windows mobile computers or PDAs. In competition with Microsoft, we can sell fully compatible single-user and multiuser ANSI-compliant DBMSs for Windows and Windows NT. Microsoft has to sell two different products: Access and SQL Server NT.
DBMS: Why change some of the product names?
McEvoy: By grouping products into families, we haven't made any products obsolete, but we're giving them family names. As part of the Sybase System 11 announcement, we'll be renaming the Watcom products to Sybase SQL Anywhere (single-user and multiuser), IQ Accelerator will be renamed Sybase IQ, and Navigation Server will be renamed Sybase MPP. The database brand is Sybase, and all of these will have the Sybase name. Watcom as a name will stay with the compilers, because it has a brand image.
DBMS: What is the main focus of the System 11 release?
McEvoy: The System 11 announcement is focused on database: SQL Server 11, Sybase IQ, Sybase MPP, and Sybase SQL Anywhere. SQL Server is the OLTP workhorse. Sybase IQ and Sybase MPP are focused on data warehousing solutions on SMP and MPP architectures. The renamed Watcom products will focus on mass deployment. Also, they will provide Transact-SQL compatibility and a developer's guide to explain how to make the same application exactly the same across the two platforms. They will also provide replication with SQL Server.
For our current SQL Server customers, SQL Server 11 is an easy upgrade. You don't have to recompile applications. The same binaries work with 100-percent compatibility. This upgrade will be for customers who need headroom from System 10 or 4.9.
In terms of OLTP, the importance is price/performance and resource usage to get the maximum number of users with good transaction rates and response times from the minimum amount of hardware. In data warehousing, Sybase IQ uses advanced indices that are optimized for decision support, instead of taking a hardware-intensive approach that uses an indexing approach designed for OLTP instead of DSS. Ad hoc queries with IQ are an order of magnitude faster, and structured queries are equally fast or faster.
DBMS: What will Transact-SQL compatibility do to Watcom's (that is, SQL Anywhere's) otherwise small footprint?
McEvoy: We'll have options. I think of them like the WordPerfect translation macros for Word, or the 1-2-3 workalike keys. Customers who require that level of compatibility, will need extra (but not significantly more) memory. A more typical configuration will be a SQL Anywhere database replicating to a Sybase SQL Server database, or running with the local database and doing remote procedure calls (RPCs) using a multitier application with PowerBuilder 5.0.
DBMS: Does Sybase IQ employ multidimensional data structures, like those you find in OLAP products?
McEvoy: In the decision-support marketplace and data warehousing, you have to differentiate between products that are optimized for preplanned queries and those that are optimized for ad hoc queries. The n-dimensional, such as Arbor's Essbase and IRI's Express, and even what Red Brick does, are really designed for when you don't know the answers but you know the questions. The advanced index approach with IQ is for the other case: Not only do you not know the answers, you don't even know what the next question will be.
With IQ's bit-wise indexing structure, you can afford to have everything fully indexed. You never have to do a table scan. With the n-dimensional approach, you're basically predefining all the paths you're going to take. You can have excellent performance, assuming you're going down one of those predefined paths. Our approach to n-dimensional is to do it with partners. All of the companies that do it can "see through" into SQL Server data.
Also, with something like Express, you have to take yet another extract from the database and load it into Express. If you have high-powered financial analysts who need to see all of these relationships, it makes sense to do the precomputes. But, you can't do that for the broad base of users and the broad set of questions that they might need to ask. The emphasis here is on ad hoc. IQ has excellent performance for preplanned queries, and it will blow the socks off anybody on the ad hoc side.
DBMS: Does Sybase IQ require specialized applications on the front end?
McEvoy: No. Any client application that uses our standard APIs will work with IQ. IQ is mated to the Sybase API, so user application programs don't have to change to use it. IQ will do the query if it can, otherwise, it passes it on to SQL Server. It uses Open Client/Open Server middleware to provide transparency. Also, it can satisfy everything through the indexes. If you're not doing OLTP updates, you don't have to have the OLTP indexes or the OLTP data. It's a very clever approach that will have a big impact on the market. The batch loads are still required. Sybase IQ doesn't do single updates. It passes them over to SQL Server.
DBMS: Can Sybase IQ do incremental batch updates?
McEvoy: Absolutely. In fact, if you've got 10 million rows and you add another million rows, IQ just builds the additional index pages. If you do something based on B-trees, you basically drop everything and rebuild the indexes. I really think this product will set a new performance standard. It will dramatically save on hardware costs. It's great if you can apply twelve processors to a select for the first user. But what happens when you add the second, the fifth, or the tenth user, and you're using all of your hardware for the first?
DBMS: How does the new Sybase IQ differ from the previous version of Expressway's IQ?
McEvoy: It will have much tighter integration with Sybase SQL Server -- not just using the same APIs, but with a much higher degree of the SQL functionality implemented for compatibility with SQL Server. More important, it's integrated completely with the SQL Server catalog, so all user authentication will come right out of SQL Server. SQL Server has catalogs, the OLTP indexes (the B-trees), and the data. What IQ adds, at the DBA's discretion, are the decision-support indexes.
DBMS: Will System 11 include object-oriented features?
McEvoy: If you look at the object problem to be solved, it's a problem that spans tools, middleware, and databases. With the acquisition of Powersoft, we now have the leading object 4GL with PowerBuilder and a leading 3GL with Watcom C++. We have a strong middleware focus, and we've had some investment and internal innovation on object databases. We've been quiet on the object front because we want to bring to the market a full object solution that solves the customers' object problem across databases, middleware, and tools, not just particular features of the database.
We won't be rolling that out as part of the System 11 announcement because we want to send clarity to the market. But you can expect some announcements in the object area later this year. We have several key people on the SQL3 committees, including Don Deutsch, Jim Melton, and Andrew Eisenberg. These people work for Sybase, so we're in the thick of this object-relational morass.
DBMS: Does your emphasis on "real-world" scalability imply that you're concerned about losing in benchmarks?
McEvoy: I don't think so. We'll have outstanding TPC-C numbers, but there's no feature in SQL Server 11 that you'd only use in a benchmark. You would use these features in your real applications. It really has been our approach all along to look at comparisons of real customer applications. You never know what the next benchmark is going to be.
The other key point here is price/performance. If you look at the Informix benchmarks, which appear in their ad, the fine print says that "in two of the three platforms, Sybase has a lower cost per transaction." And we've had the best price/performance on all platforms with one to four processors. We'll have the best price/performance at each SMP level.
DBMS: Will you beat Microsoft SQL Server on NT?
McEvoy: Yes. About half of the SQL Server 11 features that we put in for performance are not in Microsoft SQL Server 6.0. Also, you're limited on NT by the scalability of NT itself. NT on a four-processor Intel system -- according to Intel and Sequent -- scales to about 2.4 or 2.5 processors worth of power on a four-processor system. Depending on which Unix system you run on a four-processor Intel box, at the operating system level you get scalability to about 3.5 to 3.9 times one processor. Sequent has the best scalability, because it has 11 years of doing SMP. The Unix vendors have more years of experience on SMP. Microsoft has the fewest. It's not a particularly strong design center for them. The result is that the operating system limits the performance of the hardware and the DBMS.
DBMS: How will System 11 achieve the performance and scalability you're talking about?
McEvoy: The first way is a via logical memory manager. Instead of having one cache, you have an infinite number of named caches. You can assign them logically to different tables. You can assign one table to a cache, and then you have an in-memory database. You can assign a group of tables that are primarily used for OLTP to one cache, and have tables that are used primarily for decision support in other caches. With these multiple gigabytes of main memory, the idea that you're going to do everything in one cache doesn't make sense anymore.
We also support 64-bit VLM (very large memory) addressing for the operating systems that support it. Our first release of SQL Server 11.0 on the Digital Unix, which is a 64-bit operating system, will support VLM. In fact, we have it running on a 3GB machine. That will be production released with this release.
Also, the original version of SQL Server was limited to 2K block sizes, and now we have tunable block sizes. For your mixed workload environment -- where you're doing both OLTP and decision support --you can optimize the block sizes. Also, we have a new concept, called data partitioning, that's transparent to the application, but lets you slice the table into multiple partitions. That lets you do parallel inserts for multiple processors. It also allows parallel loads into the table.
All of these features are transparent to the application program. Some of them happen automatically, so the performance out of the box is better. Some of them require tuning by your DBA. With our SQL Server Manager and SQL Server Monitor, we'll provide the control mechanisms and analysis tools, such as a cache analysis tool built into SQL Server Monitor. These will provide the knobs for the DBA, and the feedback loop that provides tuning information.
DBMS: Can you quantify the performance gains?
McEvoy: For a Federal Express customer-inquiry order-tracking application, we're seeing loading times decrease by a factor of five. For this application, it's important because FedEx is loading data constantly during the day. Table scans improve by a factor of two and a half. The full application shows three times greater throughput for bread-and-butter transactions on the same two-processor SPARC system.
We just completed another application with MCI on an eight-way IBM/AIX system that is showing four-time increases on table scans, and four- to five- time decreases on loading. So we're getting verification from multiple applications that these [improvements] are what customers will see.
DBMS: I read news reports that System 11 would be delayed because of new quality controls at Sybase. How have the processes changed?
McEvoy: We've changed the development rollout process, the development process, and the customer rollout process for SQL Server 11. If you look at the way things are done in the industry, you complete your development, and, after you get things integrated, you go into an internal Alpha phase and do an initial QA. When you reach a specified quality level, you perform full regression testing to make sure you haven't broken anything from prior releases. You perform your performance testing on customer applications and industry benchmarks, and you run customer applications at the customer site to find bugs.
DBMS: If you're Microsoft, you charge your customers to test your softwareż
McEvoy: And if you're Microsoft, you keep inserting features during testing, and you complete your development. Customers are wise to this, so when you go to general availability (GA) some customers who have a bleeding need will put it into production and everybody else will wait for the first maintenance release. At Sybase, we've reengineered our development process to move everything at least one step farther back in the process, to get the bugs closer to the engineers and farther away from the customers. That's where they belong. It benefits the customers, and it also benefits us because it costs 10 times less to fix a bug at the earlier stage.
During the development phase, we've been running the regression test on the nightly builds. When we got to feature-complete, we did our performance testing on customer applications. We've also performed unaudited benchmarks during the alpha phase. We are entering beta this week, so we'll continue the regression testing. Also during the beta phase, we will go live on Sybase's most important core systems on the beta version of System 11. We plan to do that approximately one month before we go to general availability. At GA we hope to have several customers going into production -- customers who are confident in the system, so we can get production references. That will for allow a faster uptake of the product within the customer base.
DBMS: How does your new rollout process relate to the ISO 9000 certification?
McEvoy: We have a set of criteria for entering beta and for entering GA. It's called an ISO contract. We can't keep our ISO certification unless we meet those requirements. For example, before going GA, we have to run our internal application for at least a week with no priority one bugs reported. We have to run our stress tester for 72 hours without failure. We can't have any priority one bugs in the software reported within two weeks of the final date. You need one final round of testing in which you don't change anything. We can't have regressions from the currently available, latest release of the software. If we put bug fixes in System 10, we must have those forward-ported into System 11. We have criteria that are part of the ISO contract that we can't violate if we want to keep our ISO certification.
ISO forces you to specify your processes, so you find things where you can improve the process. Because ISO audits the process, you've got external verification that you're following your processes, and you've got external input to increase the efficiency of your processes and the quality of the results. We're seeing a real payoff.
DBMS: What impact did the ISO 9000 process have on the schedule?
McEvoy: From the time I took over (about seven months ago), we adjusted the schedule by about 45 days. That was a change that we made in the development process, where we kept the code line stable as we were integrating features. In other words, we were running the regression test library and we couldn't check in a new feature until you had reached stability from the prior check-in. We didn't know how long it would take. We made a prediction about that, and we were off by about 30 to 45 days, so we adjusted the schedule to accommodate that. ISO itself hasn't slowed this down. Following the processes that give quality early and result in outstanding quality when you ship, resulted in a 45-day delay against our original prediction.
DBMS: What is the timing for everything?
McEvoy: SQL Server 11 goes into beta this week (August 1) and will be released for general availability in the fourth quarter of this year, with an early partner release in October. We'll have both end users and partners in the initial betas, and many additional partners in October. Initially, SQL Server 11 will ship on five platforms: Digital Unix, HP-UX, IBM AIX, Sun Solaris, and Windows NT.
Sybase IQ is in beta, and will be released for general availability in the fourth quarter. It will be compatible with both SQL Server 10 and SQL Server 11. IQ will run on Sun Solaris, IBM AIX, Windows NT, AT&T GIS, and HP-UX.
Sybase MPP, formerly called Navigation Server, is shipping today on the AT&T platform, and will ship on at least one additional platform in the fourth quarter of this year. We're porting Sybase MPP to IBM SP2, HP, and Sun. Our goal is to ship them all in the fourth quarter of this year or the first quarter of 1996.
What was to be called Watcom 5 will be called Sybase SQL Anywhere, and it is scheduled for fourth quarter release. Our current product is Watcom SQL 4.0, and we'll announce the rebranding with the System 11 announcement. Whether that will be called version 11 or 5 -- we still have to work out those details. Sybase SQL Anywhere will support DOS, Windows, Windows NT, and Windows 95.
DBMS: Navigation Server was developed jointly by Sybase and AT&T. I understand you are bringing it over to Sybase completely?
McEvoy: Right. We're in transition now, but it's not complete yet. We're ramping up the resources and AT&T is ramping down the resources, so that by the fourth quarter, Sybase will have full development and support responsibility for the product. We're also in the process of porting it to additional hardware platforms.
AT&T has other software platforms on its hardware. We have other hardware partners for the software. Now, we'll get it back to a very regular treatment of the product. We're still leveraging the R&D that AT&T did, and there are financial arrangements associated with that, but it has had no effect on the forward-engineering of the product.
DBMS: What features are planned for later releases?
McEvoy: For SQL Server, the 11.1 release has always been the target for adding the internal parallelization for SMP platforms for improved decision-support processing. The 11.1 release is in parallel development -- much like the way Intel put parallel teams on the 486 and the 586, then the 486 team jumped to the P6. We're doing the same thing, so the 11.1 team is targeted for release about six months after general availability of SQL Server 11.0.
There are several other features to include. The focus of release 11.0 in SQL Server is SMP scalability. The focus for release 11.1 is parallelization for improved decision support.
DBMS: I've read reports that row-level locking will not be included in System 11.
McEvoy: We're doing multiple releases in parallel, 11.0, 11.1, and 11.2. Row-level locking was never targeted for 11.0. Row-level locking was targeted for late 1996. When we have that feature in the system, it will provide the option of page-level or row-level locking, much like DB2. As you know, DB2 was a page-locking system until 4.0, which I think was just released this past quarter. DB2 had a page-locking system for 11 years, from 1984 to 1995, and there were plenty of scalable applications built on that. If you design your applications to presuppose row-level locking, then you won't have high concurrency when you go to a page locking system. That's the issue with that one particular SAP application. We're working diligently on putting that feature in, but it's not in the 11.0 release.
DBMS: Why should a customer choose Sybase over Oracle?
McEvoy: We provide solutions for customers -- with databases, middleware, and tools -- for their specific classes of applications, but we do it in an open environment with products optimized for OLTP, DSS, and mass deployment. Our tools work with every database. Our middleware is designed to work with every database. Our database works with multiple tools and other middleware. We believe it's a heterogeneous world, and we don't believe that Sybase has to provide 100 percent of the solutions to customers' problems. We want to give the customer choiceżbest of breed products, but choice.
If you want Oracle tools, you have to buy an Oracle database. If you want Oracle middleware, you have to buy an Oracle database. So it's a very database-centric approach, and it offers no choice but one-stop shopping.
We believe that the mainframe will be around for quite some time. There's going to be Oracle data, Informix data, and DB2 data. An example of this is a recent procurement, where a customer had to replicate between DB2 and Oracle. Neither Oracle nor IBM could provide a solution. Sybase demonstrated a solution. Our DB2 product is in general availability, and our Oracle product is either just finishing beta or entering general availability. We blew their socks off. There was no Sybase data there, but it's heterogeneous replication.
Our value proposition is to give customers a choice, work with partners, but provide a total solution if that's what customers want to buy. And that goes beyond just software. It involves professional services, if customers would like to buy them from us. It involves architectural consulting and business process reengineering. If the customer wants to work with partners, that's fantastic too.