Computer benchmarks are like the miles-per-gallon stickers dealerships paste to the windows of new cars. There's an implied disclaimer: "Your mileage may vary." In addition, there are so many computer benchmarks--both industry-standard benchmarks and vendor-specific benchmarks--that it's often confusing for a potential server customer to know exactly what is being measured.
Therefore, it was with some trepidation that I looked at the new "three-in-one" iSeries benchmark that IBM recently announced. What exactly is IBM measuring? How relevant is this benchmark to the real-world challenges that face servers? Is there substance behind IBM's claims about the i810, or is IBM really calculating how many angels can dance on the head of microprocessor?
Needless to say, the iSeries does require some special consideration when you try to compare its horsepower to other servers. With other machines, you match apples to apples and oranges to oranges to understand how a particular server will respond to a given workload. But when you compare this machine's capabilities to the advantages or limitations of other architectures, it's like comparing bowls of fruit: The architecture of the iSeries makes it extremely resilient to a variety of different environments, and its use of storage pools and logical partitions--combined with its much-heralded performance-tuning capabilities--makes the machine one of the best all-round values for the medium-sized organization that has a real heterogeneous workload.
Still, does the iSeries really need a machine-specific benchmark to separate it from other servers in the marketplace? And why a "three-in-one" benchmark? What does it mean?
As Easy as "One, Two, Three": Java, Domino, and WebSphere
IBM's benchmark tests were performed on the recently announced iSeries model 810-2469 with feature code 7428. (The feature code 7428 indicates that the 5250 OLTP CPW rating for the system is 0 [zero]. With a rating of 0, enough interactive processing is available for a single 5250 job to perform system administration functions.) This model had a two-way processor that was running at 750 MHz and was based upon SStar technology. The system was configured with 8 GB of main storage and 30 disk arms. The disk drives were 10 K rpm with 17 GB of capacity each and were configured for RAID5 disk protection. This is a substantial system, composed of a lot of new hardware and, no doubt, chosen as the "ideal" iSeries system, based upon the current iSeries technology--stripped of peripherals but loaded with capacity.
IBM chose its three application suites for its three-in-one benchmark to showcase the capabilities of the machine as well: one for Java serving, one for Domino serving, and one for WebSphere serving. The base scenario for this benchmark is the result of running three applications simultaneously to drive the system at approximately 70% CPU utilization. IBM believes this scenario represents a "real-world data processing environment" consisting of online transaction processing (OLTP), collaboration, and Web serving. It helps that this combination represents the three environments that the iSeries has been tailored to address--three environments in which IBM has substantial expertise in all aspects of development.
jBOB OLTP Application
For the OLTP application workload, IBM used an application that conforms to the Java Business Object Benchmark (jBOB), which is a complete environment where users execute transactions against a database. The jBOB users execute a Java application that accesses DB2 database files via Java Database Connectivity (JDBC) to process transactions. The workload is centered around the transactions of an order-entry environment. These transactions include entering and delivering orders, recording payments, checking the status of orders, and monitoring the level of stock at the warehouses. The workload used a database configuration that represented 100 warehouses.
iNotes and Sametime Collaborative Application
The collaborative processing benchmark consisted of two components, iNotes Web Access users and Sametime Instant Messaging users. The iNotes Web Access simulation used the NotesBench R6i Notes workload. (IBM says that the results of this test have been audited by KMDS, Inc.) The iNotes Web Access users read, created, sent, and deleted mail and performed calendar and scheduling operations. The users accessed their 16 MB mail databases per the benchmark requirements of NotesBench. The second component of the collaborative processing benchmark simulated the actions of Sametime Instant Messaging users, sending and receiving messages against a large, 50-person buddy list.
WebSphere Web Serving Component
For the final component of the three-in-one benchmark, IBM used the Trade2 workload benchmark to simulate Web server processes. According to IBM, Trade2 can be configured to use JDBC directly or to use Enterprise Java Beans (EJBs) under WebSphere, but IBM used the JDBC mode in conjunction with WebSphere Express 5.0 without EJBs. According to IBM, the JSPs and servlets used by Trade2 were relatively simple in structure and each Web serving transaction used an average of two JSPs and two servlets. File sizes of the JSPs were 10 K or less.
No "Jack" out of the Box
IBM says it didn't mess with the configuration of the machine or the software. The three-in-one benchmark used the default settings (out-of-the-box configuration) of the iSeries, with no special tuning of storage pools, job priorities, or other attributes associated with the job execution.
- All of the applications were executed out of the same memory or main storage pool (*BASE) and shared the same processor resources. OS/400 Logical Partitioning (LPAR) was not used.
- QSECURITY was set to level 40 (requiring more authorization checking).
- Secure Sockets Layer (SSL) processing was added to the Trade2 benchmark. This makes a user log off and log in again as a new user, requiring a "full handshake" process for new transactions.
- Global security was enabled in WebSphere Express using the OS to validate users.
Two separate Domino server partitions were used for the iNotes Web Access and Sametime Instant Messaging processing. (IBM says that both servers could have been supported by a single Domino server, using fewer resources, but the choice to use two was based on how real customers typically configure the box.)
The Domino Directory (name and address book) consisted of 6,000 persons.
In other words, IBM chose the most conservative, plain vanilla configurations, insisting that neither the machine nor the software was altered to meet performance expectations. This is a fairly common statement among hardware vendors, but recent evidence shows that most other vendors go deep under the covers to maximize either the software or the hardware configurations.
How Ya Doin'?
So, on the surface, the specifications for the iSeries three-in-one benchmark look very impressive. They are, indeed, very close to a real customer environment--though perhaps at a configuration level that is a little crude. (Any organization that would invest in an i810 and these applications suites would certainly play a bit with the tuning of both hardware and software.)
But the rather amazing thing about the benchmark was the i810's performance results. When IBM pushed the system to handle 2,850 users at a resource consumption level of 96.8%, the machine was still able to maintain sub-second response time for all applications: OLTP, collaboration, and Web serving. (See Figure 1.)
Average CPU Utilization
Average Response Time and/or Throughput
Number of Users
Lotus Domino Web Access and Lotus Instant Messaging
0.509 seconds (109.3 trans/second)
| || |
| || |
Figure 1: IBM pushes the envelope.
What's It Mean?
The three-in-one benchmark really showcases the power of the iSeries architecture and the resiliency of its hardware. Its multi-environment capabilities--bringing all these environments into a single footprint--really dwarfs the capacities of its competitors. All of these applications seem to run--out of the box--at optimum levels, without degrading the performance of any individual member.
So what does this mean to iSeries customers?
It means they can be confident that the architecture and the hardware are up to the challenges of real-world environments with a pretty fantastic performance level. It also means that the iSeries is establishing a new standard for server cost-effectiveness.
The real problem with the three-in-one benchmark is not the equipment, the configuration, or the applications that are running. The real problem challenging IBM is convincing new non-iSeries customers to look at the results. Unfortunately, IT management rarely considers real technical specifications when choosing between server architectures anymore, instead looking at the short-term bottom line or accepting whatever black box is coming off the Business Partner's quota sheet. Companies don't buy servers these days: They buy server packages built around software suites, and the truth is, it's much easier for a salesman to bundle in a lesser server than to convince a customer that money can be saved through server consolidation.
Yet perhaps that's the gold lining of the three-in-one benchmark: For once, BPs will have the opportunity to show customers how they can bring all their applications into a single multi-environment footprint, maintain supreme performance, and still save a bundle.
Good work, IBM.