Performance of environments using DB2 Connect Enterprise ...public.dhe.ibm.com/software/dw/linux390/perf/ZSW03032-USEN-00.pdfThe objectives of this project were to analyze the performance

Performance of environments using DB2 Connect Enterprise Edition

1


February 2008


2

Table of Contents

Objective .................................................................................................................................3 Executive summary.................................................................................................................3 Summary.................................................................................................................................4 Hardware equipment and software environment.....................................................................6

Host hardware and software ...............................................................................................6 Network setup.................................................................................................................6 Storage server setup.......................................................................................................6 z/VM Guest Setup...........................................................................................................7 Client hardware...............................................................................................................7

Software..............................................................................................................................7 Test environment ....................................................................................................................8

Trade 6................................................................................................................................9 Trade 6 topology overview................................................................................................10 IRWW................................................................................................................................10

Setup.....................................................................................................................................11 DB2 Connect - connection pooling and concentrator function ..........................................12

DB2 Connect parameters .............................................................................................12 DB2 for z/OS (DSNZPARM parameter) ........................................................................13 Enable the DB2 Connect connection concentrator function..........................................14

Workload management policy for z/OS.............................................................................15 WebSphere Application Server setup ...............................................................................15

Connection monitoring and CPU utilization charts explained................................................16 Monitor active connections................................................................................................16 CPU utilization charts explained .......................................................................................17

Results ..................................................................................................................................18 Considerations for all scenarios ........................................................................................18 Parameters common for all test cases..............................................................................18 Trade 6 results ..................................................................................................................18 Scaling DB2 Connect memory ..........................................................................................19 Scaling DB2 Connect virtual CPUs ...................................................................................20 Scaling DB2 Connect parameter MAXAGENTS ...............................................................21 Compare DB2 Connect to non-DB2 Connect....................................................................23

IRWW results ........................................................................................................................24 Scaling DB2 Connect memory ..........................................................................................25 Scaling DB2 Connect CPUs..............................................................................................26 Scaling DB2 Connect MAXAGENTS ................................................................................27 Comparison of SQLJ versus JDBC...................................................................................29 Compare DB2 Connect to non-DB2 Connect....................................................................30

Appendix A. Detailed set up examples..................................................................................31 DB2 Connect configuration commands for Trade 6 ..........................................................31 Buffer pools used with the Trade 6 database on z/OS......................................................33 DB2 Connect configuration commands for IRWW ............................................................33 Buffer pools used with the IRWW database on z/OS........................................................34

Appendix B. Other Sources of Information............................................................................34


3

Objective1

DB2 Connect™ Enterprise Edition V9.1 for Linux® on zSeries® (DB2 Connect) connects LAN-based systems and their applications to the company's mainframe host DB2® databases. Additionally, DB2 Connect can be used in a three tier environment and can act as a gateway concentrating large numbers of SQL connections from various clients to a fewer, well-defined number of connections to the DB2 database on z/OS®.

The focus of this report was on the connection concentrator functions of DB2 Connect. This feature enables a predictable, controlled load on the database on z/OS from a DB2 Connect server and may reduce mainframe resource usage as fewer DB2 threads may be defined. The capacity of active connections from the DB2 Connect gateway to the z/OS DB2 database was dependent on the resources available on the DB2 Connect gateway. The scaling behavior from DB2 Connect for the number of CPUs, memory, and various connection concentration levels were examined.

DB2 Connect was hosted on Linux for System z™, which was implemented together with the WebSphere® Application Server environment as z/VM® guests. This setup provided an integrated solution with DB2 Connect providing gateway and connection concentration for a three tier environment with the Trade 6 workload and gateway and connection concentration in a two tier environment with the IBM Relational Warehouse Workload (IRWW). The two application workloads, Trade 6 and IRWW, with their different architectures were tested to examine how application implementation characteristics might affect settings for optimal performance.

The objectives of this project were to analyze the performance of DB2 Connect environments. The information in this paper should not be used to develop capacity sizing rules-of-thumb for customer environments. For sizing and capacity planning, contact our specialists from the IBM System z sizing team.

Executive summary

This paper shows the results from our performance measurements with environments using DB2 Connect Enterprise Edition. For our workload, a DB2 Connect system with 1 GB of main memory and up to 2 virtual CPUs was suitable for all cases. The use of the connection concentrator function of DB2 Connect was a suitable tool to control the load on the DB2 database, but it requires specific prerequisites on DB2 and WebSphere Application Server

1 This paper is intended to provide information regarding performance of environments using DB2 Connect Enterprise Edition. It discusses findings based on configurations that were created and tested under laboratory conditions. These findings may not be realized in all customer environments, and implementation in such environments may require additional steps, configurations, and performance analysis. The information herein is provided “AS IS“ with no warranties, express or implied. This information does not constitute a specification or form part of the warranty for any IBM products.


4

settings as well as the characteristics of the workload itself. The impact of DB2 Connect on the throughput in the three tier environment was very small. It was higher in the two tier environment. Additionally, we identified the Workload Manager (WLM) settings on z/OS as an important parameter for database performance.

Summary

This chapter provides a short summary of our test results. Our test results and recommendations are specific to our environment. Parameters useful in our environment might be useful in other environments, but are dependent on application usage and system configuration. You will need to determine what works best for your environment. For our detailed test results information, see Results.

The following are our summary results:

General

We identified the WLM settings on z/OS as an important parameter for

database performance.

Limiting the number of connections to DB2 on z/OS can be used to limit

the load created from this source on DB2. This protects the database

against load peaks from the workload and improves mainframe

resource utilization.

The DB2 Connect server may sustain continuous 100% CPU utilization

with improving throughput even while its connections to z/OS DB2 grow

as much as 12X.

Connection concentration

We identified several prerequisites for the mode where connections were concentrated on the DB2 Connect system.

On the DB2 Connect system, the parameter, MAX_CONNECTIONS,

must be at least 1 greater than MAXAGENTS.

On the z/OS system the DSNZPARM parameter, MAXDBATS, must be

equal to or greater than MAXAGENTS.

On the WebSphere Application Server level, the parameter,

resultSetHoldability, must be set to 2.

The application must have only a small amount of long running

transactions (ideally none) and all cursors must be closed after commit.

Trade workload

A DB2 Connect guest with 512 MB of main memory started swapping

slightly with 400 Trade users. Our maximum number of Trade users of


5

700 was well supported with a DB2 Connect system with 1 GB of

memory. An additional 1 GB of main memory provided no advantage. It

seems that about 512 MB of main memory is needed for each 400 users.

We did all further tests with 1 GB of main memory for the DB2 Connect

system.

The DB2 Connect system with a single CPU was overloaded with 400

concurrent users with this workload. It is recommended that DB2

Connect be run with a minimum of two CPUs when a high volume of

concurrent active clients are likely.

When controlling the load on the DB2 on z/OS created from 600 Trade

users by scaling the numbers of active connections via DB2 Connect

MAXAGENTS from 50 to 600, a factor of 12X, the load on the DB2 on

z/OS only varied by a factor of 2x. This shows that DB2 Connect was a

suitable tool to control the load on the DB2 database.

When the Trade 6 transactions are routed though DB2 Connect,

throughput degrades up to 8%. This may be considered a small impact

to achieve predictable, efficient utilization of mainframe DB2 resources

while handling widely varying client access volumes.

IRWW workload

DB2 Connect memory size does not affect IRWW throughput in these

scenarios. A memory size of 1 GB was selected for the remainder of the

IRWW test cases.

The IRWW workload throughput is not affected by the number of CPUs.

A single CPU for the DB2 Connect system would be sufficient for this

type of workload, but for a better comparison with the Trade 6 results,

we decided to use two CPUs on the DB2 Connect system for further

runs.

For the think time values used, a MAXAGENTS value of 100 was

sufficient for this workload. This demonstrates that limiting connections

to DB2 on z/OS can be done without a degradation of the throughput,

up to a certain load level.

We selected the SQLJ workload driver because, in our environment, it

produces slightly higher throughput at lower CPU utilization than the

JDBC driver.

The IRWW workload creates a significant lower load on the system than

the Trade workload. DB2 Connect has here a higher impact on

throughput than in the Trade 6 workload case. Bypassing DB2 Connect

resulted in an improvement between 6% and 30%. It seems that

WebSphere balances the request rate in a way that the pressure on

DB2 Connect is lower.


6

Hardware equipment and software environment

This chapter provides details on the hardware and software used in our testing. Topics include:

Host and client hardware used

Host and client software used

Network setup

Storage server setup

z/VM guest setup

The test environment

Workload description

Host hardware and software

The following section lists the hardware, software, network, and configurations we used for the environment on System z.

Hardware

Two LPARs on a 16-way IBM System z9®, type 2094-S18, equipped with:

Table 1. System memory and CPUs

System/Guest Memory Dedicated CPUs

z/VM 16384 MB (4096 MB expanded) 5 or 6

z/OS 8192 MB 4 - 7

Additionally, the z/VM LPAR used an OSA-Express2 gigabit Ethernet card (OSA code level 0805)

Network setup

Our network setup consisted of the following: The clients were connected to the z/VM LPAR via Fiber Gigabit Ethernet

Interface

The z/VM guests used virtual guest LANs

The z/VM LPAR and the z/OS LPAR were connected via a

HiperSockets™ connection

Storage server setup

Enterprise Storage Server® (ESS) 2105 800

IBM 3390 disk model 3

Physical DDMs with 15,000 RPMs


7

z/VM Guest Setup

Table 2 shows the z/VM guest configuration we used for our tests. We used this configuration for our tests using the Trade 6 workload. For tests using the IRWW workload the WebSphere Application Server guests were not required.

Table 2. Test system configuration: WebSphere Application Server

System/Guest Memory # of Virtual

CPUs

DB2 Connect 512 MB (1024 MB) (2048 MB) 1 (2)

WebSphere Application Server (1) 2048 MB 2




Client hardware

1 x330 PC with 2 processors, 1.26 GHz

Software

Table 3. Host and client software used

Product Version/Level

WebSphere Application Server 6.1.0.3, Build Number: cf30646.29

DB2 Connect Enterprise Edition for Linux on zSeries

9.1

Red Hat Enterprise Linux RHEL 4 ES (client)

WebSphere Studio Workload Simulator iwl-0-03309L

SUSE Linux Enterprise Server SLES9 SP3, kernel level 2.65-7.244-

s390x (all z/VM guests)

DB2 for z/OS 9.1

z/OS 1.8

z/VM 5.2


8

Test environment

Figure 1 shows the setup of the Trade 6 DB2 Connect environment

Figure 1. Trade 6 - DB2 Connect test environment Figure 2 shows the setup of the IRWW DB2 Connect environment.

Figure 2. IRWW-DB2 Connect test environment


9

The test environment consisted of an IBM System z server and an IBM System x™ server. The System z contained a z/VM LPAR with five guests and a z/OS LPAR. The network was split as follows:

The System x and the z/VM LPAR on the System z were connected via a

1 GB Ethernet connection. The System x contained the Trade 6 or IRWW

workload generator, which generated the workload.

The network on z/VM was implemented as a guest LAN type of

HiperSockets, which was used by all guests.

The z/OS LPAR was connected to the z/VM LPAR through a

HiperSockets connection and contained the DB2 for z/OS server.

Workload description

Trade 6

The IBM Trade Performance Benchmark Sample for WebSphere Application Server (otherwise known as Trade 6) is the fourth generation of the WebSphere end-to-end benchmark and performance sample application. The Trade benchmark is designed and developed to cover the significantly expanding programming model and performance technologies associated with WebSphere Application Server. This application provides a real-world workload, enabling performance research and verification tests of the Java™ 2 Platform, Enterprise Edition (J2EE) 1.4 implementation in WebSphere Application Server, including key performance components and features.

Overall, the Trade application is primarily used for performance research on a wide range of software components and platforms. The latest revision of Trade builds off of Trade 3, by moving from the J2EE 1.3 programming model to the J2EE 1.4 model that is supported by WebSphere Application Server V6.0. Trade 6 adds DistributedMap based data caching in addition to the command bean caching that is used in Trade 3. Otherwise, the implementation and workflow of the Trade application remains unchanged.

Trade's new design enables performance research on J2EE 1.4 including the new Enterprise JavaBeans™ (EJB™) 2.1 component architecture, message-driven beans, transactions (1-phase, 2-phase commit) and Web services (SOAP, WSDL, JAX-RPC, enterprise Web services). Trade 6 also drives key WebSphere Application Server performance components such as dynamic caching, WebSphere Edge Server, and EJB caching.


10

Figure 3 shows the J2EE components that make up the Trade application.

Trade 6 topology overview

Trade provides two server implementations of the emulated Trade brokerage services.

EJB - Database access uses EJB 2.1 technology to drive transactional

trading operations.

Direct - Uses database and messaging access through direct JDBC

and JMS code.

The J2EE programming model provides managed, object-based EJB components. The EJB container provides declarative services for these components such as persistence, transactions, and security. The J2EE programming model also supports low-level APIs such as JDBC and JMS. These APIs provide direct access to resource managers such as database and message servers.

For this test environment we used only the Direct mode.

For the tests described in this paper the Trade 6 database was populated with 4000 user IDs and 2000 stock quotes.

IRWW

IRWW is an OLTP workload that consists of seven transactions. Each transaction consists of one to many SQL statements, each performing a distinct business function in a predefined mix.


11

The seven transaction types, a brief description of each, and the percentage of the transaction mix follows:

Neworder

Performs various SELECTS, FETCHES, UPDATES, and INSERTS in support of the receipt of new customer orders and runs as 22% of the total transaction mix.

Order Status

Performs various SELECTS and FETCHES in support of providing the status of an order and runs as 24% of the total transaction mix.

Payment

Performs SELECTS, FETCHES, UPDATES, and INSERTS in support of received customer payments and runs at 22% of the total transaction mix.

Price Change

Performs an UPDATE in support of changing the price of an item and runs as 1% of the total transaction mix.

Price Quote

Performs various SELECTS in support of providing the price of a set of items and runs as 25% of the total transaction mix.

Stock Level

Performs a JOIN and various SELECTS in support of providing the current stock level of an item and runs at 4% of the mix.

Delivery

Performs various SELECTS, UPDATES, and DELETES in support of the delivery of a group of orders and runs as 2% of the total transaction mix.

The IRWW database that we built contained 8 inventory stock warehouses and 10 sales districts

Setup

This chapter includes the following setup topics:

DB2 Connect connection pooling and concentrator function

Workload management policy for z/OS

WebSphere Application Server setup


12

DB2 Connect - connection pooling and concentrator function

The following is a short description of the DB2 Connect parameters that are configurable and relate to connection pooling and concentration.

The parameters described in this section are configuration parameters and are specified using the following command:

db2 update dbm cfg using <parameter> <value>

Connection Pooling

Connection Pooling is a simple technique that allows reuse of an established connection infrastructure for subsequent connections and handles connection volume and helps to reduce the overhead of database connections. It avoids the overhead of opening and closing connections by holding the connections active in a pool. You can configure the pool size using the NUM_POOLAGENTS configuration parameter.

Connection Concentrator

The Connection Concentrator enables a fewer number of threads in z/OS DB2 and thus lower associated hardware usage by DB2 for z/OS to service the client requests created from that DB2 Connect server (see also http://www.ibm.com/developerworks/db2/library/techarticle/dm-0503katsnelson/index.html ). A higher number of clients are connected to the DB2 Connect server, which is really connected to the database, rather than the clients connected directly to z/OS DB2. It introduces the concept of agents as follows:

Logical Agent (LA) - handles user context

Coordinating Agent (CA) - owns DB2 connections and processes

Any time an application user connects, DB2 Connect assigns a logical agent. A coordinating agent is needed to pass SQL to z/OS DB2, so one is assigned as soon as a transaction is initiated. The coordinating agent is disassociated from the logical agent and is returned to the pool when the transaction is finished. The term logical agent was created to try to explain the decoupling of the application state from agents that service the work when connection concentration is enabled. A logical agent is just a concept. There is no agent involved. A logical agent represents the state information associated with the application connection, which is then tied to the agent processing the transaction.

DB2 Connect parameters

MAXAGENTS - The maximum number of DB2 Connect clients that can concurrently execute work against DB2 for z/OS is defined by the MAXAGENTS value. This value should be less than or equal to the MAXDBATS value on z/OS.


13

Coordinating Agents - The configuration parameter MAX_COORDAGENTS limits the number of coordinator agents. A co-coordinator agent (COORD agent) is the agent that handles the connections to the database. The setting of MAXAGENTS controls the total number of COORD + sub-agents (and a few other types of agents) that can actually be spawned by the DB2 Connect instance. However, because a pure DB2 Connect instance will not have any sub-agents, MAX_COORDAGENTS and MAXAGENTS are the same thing. In other words, even if MAXAGENTS is set higher than MAX_COORDAGENTS, in a pure DB2 Connect server environment, the number of agents used will never exceed the MAX_COORDAGENTS.

For each run we set MAX_COORDAGENTS to be the same as MAXAGENTS.

MAX_CONNECTIONS - The configuration parameter, MAX_CONNECTIONS, controls the number of connections allowed to the DB2 Connect server from the client side. When MAX_CONNECTIONS is larger than MAXAGENTS, the system runs in connection concentrator mode.

NUM_POOLAGENTS - The configuration parameter, NUM_POOLAGENTS, indicates the number of agents that, when not assigned work, will be kept active (pooled).

For each run we set NUM_POOLAGENTS to be the same as MAXAGENTS.

NUM_INITAGENTS - The configuration parameter, NUM_INITAGENTS, indicates the number of idle agents spawned at db2start. These agents are like any agent and can become COORD, sub-agents, etc. as required.

Because NUM_INITAGENTS just primes the agent pool with idle agents, its value should not exceed NUM_POOLAGENTS.

The DB2 command db2 get dbm cfg shows the default settings for DB2 after installation of DB2 Connect for the above parameters. An example is shown below.

Max number of existing agents (MAXAGENTS) = 200 Agent pool size (NUM_POOLAGENTS) = 100(calculated) Initial number of agents in pool (NUM_INITAGENTS) = 0 Max number of coordinating agents (MAX_COORDAGENTS) = MAXAGENTS Max no. of concurrent coordinating agents (MAXCAGENTS) = MAX_COORDAGENTS Max number of client connections (MAX_CONNECTIONS) = MAX_COORDAGENTS

DB2 for z/OS (DSNZPARM parameter)

On our DB2 for z/OS system we used the DSNZPARM parameter MAXDBATS.


14

MAXDBATS - The maximum number of database access agents from the DB2 connection pool. You can do a display command on DB2 for z/OS to see how many connections are concurrently active.

The z/OS for DB2 command to display this information is:

-dis ddf detail

See Monitor active connections for more details on this output.

From the results of this command, you can decide if MAXDBATS and/or MAXAGENTS need to be increased. MAXDBATS should be equal to or greater than MAXAGENTS. For instance, if MAXAGENTS is 100 and MAXDBATS is 200 and you see 100 active connections from the display command, you will know the DB2 Connect connection concentration pool has limited the number of connections, while the DB2 connection pool can service 100 more connections from other sources.

Enable the DB2 Connect connection concentrator function

This section describes the requirements to enable the DB2 Connect connection concentrator on the DB2 Connect system.

On the DB2 Connect system, the following must be done:

MAX_CONNECTIONS must be at least 1 greater than MAXAGENTS

(Number of logical agents greater than the number of coordinating

agents).

MAXDBATS on the z/OS system must be equal to or greater than

MAXAGENTS.

MAX_COORDAGENTS controls the number of inbound connections

active at any time. (Equals the number of logical agents.) Relation:

MAXAGENTS < MAX_CONNECTIONS < 64,000

On the WebSphere Application Server level, the following must be done:

Set the parameter resultSetHoldability=2. For more information see

WebSphere Application Server setup.

On the Application level, the following must be done:

DB2 Connect can dispatch agents to another client connection only after a commit and when all SQL cursors are closed. This requires that the application:

has only a small amount of long running transactions (ideally none)

closes all cursors after commit


15

Workload management policy for z/OS

It is important to set the proper workload management policy for DB2's Distributed Data Facility (DDF). The default service classification for DDF is discretionary. This service class is the second lowest available and means that once the system becomes busy, all distributed DDF work will run at a very low priority.

The following service classes were defined for DDF and to the DB2 address space started tasks:

SYSSTC

Built in service class. Used for DB91IRLM. High priority service class. Only 'SYSTEM' service class is higher.

DB2ADDRS

Service class for DB91MSTR, DB91DBM1, and DB91DIST. Uses importance=1, velocity=80. Slightly lower than the IRLM address space.

DDFWORK

Service class for DDF. Uses importance=2, velocity=80. Slightly lower priority than the DB2 address spaces.

For information about Workload Management and defining goals through the service definition see z/OS MVS™ Planning: Workload Management.

For general information about DB2 for z/OS see DB2 for z/OS Performance Monitoring and Tuning Guide.

WebSphere Application Server setup

The data source definition in WebSphere for the Trade 6 database needs the following changes to the connection pool:

Create a new attribute in the connection pool properties as

resultSetHoldability and set the value to 2.

This setting controls the cursor behavior when committing a transaction. The possible values are:

1 = Hold cursors at commit

2 = Close cursors at commit


16

Connection monitoring and CPU utilization charts explained

This chapter contains information on how we monitored our active connections as well as a sample CPU utilization chart.

Monitor active connections

The following information describes how we monitored the connections.

WebSphere - In the administrative console, we clicked Monitoring and Tuning > Performance Viewer > Current activity. We then checked the following fields:

Connection pools - The connection pool summary lists all data source

connections that are defined in the application server and shows their

usage over time.

Thread Pools - The thread pool summary shows the usage of all thread

pools in the application server over time.

DB2 Connect - On the DB2 Connect server guest we ran a script that counted the number of database agent processes running every 30 seconds during the test.

z/OS - The z/OS for DB2 command to display the active connection is -dis ddf detail. An example of the output produced by this command is shown below.

Note:

Bold is added for emphasis only.

DSNL080I -DB91 DSNLTDDF DISPLAY DDF REPORT FOLLOWS: DSNL081I STATUS=STARTD DSNL082I LOCATION LUNAME GENERICLU DSNL083I DB91ZOS USIBMT6.DB91ZOS -NONE DSNL084I TCPPORT=446 SECPORT=0 RESPORT=447 IPNAME=-NONE DSNL085I IPADDR=::9.12.22.95 DSNL086I SQL DOMAIN=lndia3.pdl.pok.ibm.com DSNL086I RESYNC DOMAIN=lndia3.pdl.pok.ibm.com DSNL090I DT=I CONDBAT= 10000 MDBAT= 1000 DSNL092I ADBAT= 198 QUEDBAT= 0 INADBAT= 0 CONQUED= 0 DSNL093I DSCDBAT= 85 INACONN= 320 DSNL099I DSNLTDDF DISPLAY DDF REPORT COMPLETE

The following fields are associated with DDF threads and connections:

condbat

Maximum number of inbound connections as determined by the "MAX REMOTE CONNECTED" value in the DSNTIPE installation panel. This value must be greater or equal to the total number of connections expected to be open at the same time, especially as the value of mdbat.


17

mdbat

Maximum number of database access threads as determined by the "MAX REMOTE ACTIVE" value in the DSNTIPE installation panel. This effectively determines the maximum number of active slots. That is, the maximum number of concurrent active database access threads that could potentially be executing SQL.

adbat

Current number of active database access threads. This value increases as new database access threads get created or become active. The value decreases as database access threads terminate or become inactive. When this value reaches or exceeds the mdbat value that is indicated in the DSNL090I message, new or inactive database access threads (DBATs), or new or inactive connections must be queued, which leads to an increasing response time. When this value exceeds the condbat value, new connection requests are rejected.

CPU utilization charts explained

In the charts in this report, the CPU utilization values are always normalized in a way that 100% means one CPU is fully utilized.

Figure 4 is an example of a CPU utilization chart for a z/OS LPAR and a z/VM LPAR in relation to z/VM guest CPU utilization

Figure 4. Sample CPU utilization chart The CPU utilization charts show the utilization for each LPAR (z/VM or z/OS) by a line with triangle symbols. The details of how the z/VM guests use the CPUs are shown with stacked bars, which should sum up to the z/VM utilization without the CP related CPU part, so it is always a little lower than the z/VM


18

LPAR utilization. If the CPU utilization from a guest in the legend is not visible, that means it is too low to display.

Results

This chapter provides our detailed test results and conclusions.

Considerations for all scenarios

You can relate for all workloads one workload user with an active network connection path from the client via WebSphere, DB2 Connect, to the DB2 on z/OS. The exception is, when DB2 Connect runs in the connection concentration mode, in that case the number of connections between DB2 Connect and DB2 are fewer than the number of clients connected to DB2 Connect which means users which do not get an connection to the database are queued within DB2 Connect.

Parameters common for all test cases

The DB2 Connect configuration parameters were set as follows, unless otherwise specified:

NUM_POOLAGENTS = MAX_COORDAGENTS = MAXCAGENTS

MAXCONNECTIONS = 1000

The DB2 for z/OS DSNZPARM parameter MAXDBATS was set to 1000.

Note:

All guest CPUs described in the following tables are virtual CPUs. The CPUs assigned to the z/VM and z/OS LPARs are physical CPUs.

Trade 6 results

The following configuration was used for all Trade 6 measurements.

Table 4. Trade 6 measurements configuration

WebSphere Application

Server 1 z/VM Guest


Server 2 z/VM Guest


Server 3 z/VM Guest


Server 4 Guest

LPARs

Memory CPUs Memory CPUs Memory CPUs Memory CPUs z/VM

CPUs

z/OS

CPUs

2 GB 2 2 GB 2 2 GB 2 2 GB 2 5 6 or 7

The WebSphere Application Server configuration parameters for the Trade 6 data source connection pool were configured with 10/100 (min/max) connection entries and the Web Connection Pool Thread size was set to 50/200 (min/max).


19

In all tests, the Trade 6 brokerage services were configured to use the "direct" method and the number of user IDs was set to 4000.

The number of stock quotes was set to 2000.

The Trade 6 workload driver (WebSphere Studio Workload Simulator) was configured for a think time of 250 milliseconds and distributed all transaction requests in a round robin fashion to each of the WebSphere Application Servers. For example, if 100 users were specified, each WebSphere server was assigned 25 users.

Scaling DB2 Connect memory

In these tests, the DB2 Connect z/VM guest used two virtual CPUs and 512 MB, 1 GB, and 2 GB of memory, depending on the run. The Trade 6 workload driver users were scaled from 50 to 600. The z/VM LPAR was assigned five CPUs and the z/OS LPAR was assigned six CPUs. The parameter, MAXAGENTS, was set high enough that the number of active connections was not limited.

Figure 5 shows throughput when scaling DB2 Connect z/VM guest memory from 512 MB to 2 GB and Trade users from 400 to 600.

Figure 5. Trade 6 memory scaling throughput


20

Observations

At 400 users, the throughput for the three memory configuration scenarios was almost identical. For the 1 GB and 2 GB memory scenarios, the throughput is also almost identical. Using more than 400 Trade users in the 512 MB scenario causes significant swapping on Linux.

Conclusion

We did not exceed 400 users at 512 MB of main memory on the DB2 Connect system. Although a relatively small amount of swapping occurs at 400 users, swapping increases above 400 users so we did not want go above that number of users.

It is recommended that if MAXAGENTS is set to 400 or greater, DB2 Connect should be run with 1 GB of memory for active connections when running applications similar to Trade 6. There is nearly no advantage to 2 GB of main memory for our test scenarios. Therefore, we used 1 GB of main memory on the DB2 Connect system for all further runs.

Scaling DB2 Connect virtual CPUs

In these tests, one or two virtual CPUs were defined for the DB2 Connect z/VM guest and 1 GB of memory was used. The number of Trade 6 users was scaled from 50 to 600. The z/VM LPAR was assigned five CPUs and the z/OS LPAR was assigned six CPUs. The parameter, MAXAGENTS, was set high enough that the number of active connections was not limited.

Figure 6 shows throughput for DB2 Connect z/VM guests with one and two virtual CPUs, scaling Trade users from 400 to 600.

Figure 6. Trade 6 workload - CPU scaling throughput


21

Figure 7 shows CPU utilization for DB2 Connect z/VM guests with one and two virtual CPUs when the Trade users were scaled from 400 - 600.

Figure 7. Trade 6 CPU scaling

Observations

With our workload, a DB2 Connect server with one virtual CPU was fully utilized quickly, which limited the throughput. Throughput in the one CPU scenario reaches its peak at 400 users and decreases beyond that. When DB2 Connect has more than one CPU, throughput improves over the one CPU case by a factor of 1.35 with 600 users.

Conclusion

DB2 Connect with a single virtual CPU is mostly overloaded with this high number of users and this workload. It works more effectively in a multi-CPU environment. It is recommended that DB2 Connect be run with a minimum of two virtual CPUs to achieve additional throughput.

Scaling DB2 Connect parameter MAXAGENTS

In this test, one virtual CPU was defined for the DB2 Connect z/VM guest and 1 GB memory was used. A constant load of 600 Trade 6 users was used. The parameter, MAXAGENTS, was scaled from 50 to 600 to scale the number of active connections between DB2 Connect and z/OS DB2. User connection requests which did not get one of the active connections were queued within DB2 Connect. We used only one virtual CPU for the DB2 Connect because this is a scenario which creates a more moderate load.

The z/VM LPAR was assigned five CPUs and the z/OS LPAR was assigned six CPUs.


22

Figure 8 shows throughput for a DB2 Connect z/VM guest with one virtual CPU, scaling MAXGENTS from 50 to 600.

Figure 8. Trade 6 MAXAGENTS scaling throughput

Figure 9 shows CPU utilization for DB2 Connect z/VM guests with one and two virtual CPUs, scaling MAXAGENTS from 50 to 600


23

Figure 9. Trade 6 workload - MAXAGENTS scaling Observations

The DB2 Connect CPU utilization is near 100% in all cases. Between 50 - 200 MAXAGENTS, throughput is flat. With a MAXAGENTS value of 300 or more, the throughput increases according to the increase in active connections and then flattens out when it reaches the case where the connections are no longer concentrated. The CPU load on z/OS behaves in the same manner.

Throughput at 400 MAXAGENTS is 65% higher than throughput at 200 MAXAGENTS, even though DB2 Connect is running at almost 100% of available CPU.

Conclusion

The DB2 Connect connection concentration is a suitable feature to control the load on the database on z/OS. Of course, when it limits the number of connections, this prevents a further increase of the throughput, once the capacity of a single connection is exceeded. In this case a MAXAGENTS value of 300 will limit the CPU load on DB2 to a maximum of half of the 6 CPUs from that source.

Compare DB2 Connect to non-DB2 Connect

In these tests, the Trade 6 workload driver sent all transaction requests directly to DB2 on z/OS, bypassing DB2 Connect. This set of results was compared to the scenario of DB2 Connect in a z/VM guest (Figure 6) using two virtual CPUs and 1 GB memory. The number of Trade 6 users was scaled from 500 to 700. The parameter, MAXAGENTS, was set high enough that the number of active connections was not limited. The z/VM LPAR was assigned five CPUs and for this test we assigned seven CPUs to the z/OS LPAR to support 700 Trade users.


24

Figure 10 shows throughput using DB2 Connect and not using DB2 Connect and scaling users from 500 to 700.

Figure 10. Trade 6 workload - DB2 Connect versus non-DB2 Connect

Observations

Throughput increases for the bypass case, which is as expected. The impact of DB2 Connect is 6% to 8%, even with 700 users.

Conclusion

When the Trade 6 transactions are routed though DB2 Connect, throughput degrades up to 8%. This may be considered a very small impact to obtain predictable thread resource utilization on z/OS DB2.

IRWW results

The IRWW workload driver was the SQLJ version, unless otherwise specified.

The number of IRWW workload threads was set to 500, unless otherwise specified.


25

Scaling DB2 Connect memory

In these tests, the DB2 Connect z/VM guest used two virtual CPUs and 512 MB, 1 GB, and 2 GB of memory, depending on the run. The number of user threads was held constant at 500. The IRWW workload driver's think time value was varied from 2000 milliseconds to 5000 milliseconds to vary the workload. The z/VM LPAR was assigned six CPUs and the z/OS LPAR was assigned four CPUs. The parameter, MAXAGENTS, was set high enough that the number of active connections was not limited.

Figure 11 shows throughput for scaling memory in a DB2 Connect z/VM guest from 512 MB to 2 GB and scaling think times from 2000 to 5000 milliseconds.

Figure 11. IRWW memory scaling throughput Observations

At a think time of 2000 milliseconds there is a slight throughput advantage for 512 MBs. At higher think times there is little throughput difference between the various memory sizes.

Conclusion

DB2 Connect memory size does not affect IRWW throughput in these scenarios. A memory size of 1 GB was selected for the remainder of the IRWW test cases.


26

Scaling DB2 Connect CPUs

In these tests, the number of virtual CPUs defined for the DB2 Connect z/VM guest was varied from one to two. The memory size was 1 GB. The think time value for the IRWW workload driver was varied from 2000 milliseconds to 5000 milliseconds and the number of user threads was held constant at 500. The z/VM LPAR was assigned six CPUs and the z/OS LPAR was assigned four CPUs. The parameter, MAXAGENTS, was set high enough that the number of active connections was not limited.

Figure 12 shows throughput for DB2 Connect z/VM guests with one and two virtual CPUs and scaling think times from 2000 to 5000 milliseconds.

Figure 12. IRWW CPU scaling throughput


27

Figure 13 shows CPU utilization for DB2 Connect z/VM guests with one and two virtual CPUs and scaling think times from 2000 to 5000 milliseconds.

Figure 13. IRWW workload - CPU scaling Observations

There is a slight throughput advantage in the one CPU case for think times of 2000 milliseconds and 2500 milliseconds. At higher think times there is no difference in throughput. CPU utilization on the DB2 Connect server for the two CPU case is slightly higher than the one CPU case. The overall CPU load on the systems is very low (be aware that 100% means one CPU is fully utilized).

Conclusion

The IRWW workload throughput is not affected by the number of virtual CPUs. A single CPU for the DB2 Connect system would be sufficient for this type of workload. For a better comparison of the DB2 Connect behavior under these two very different workloads, we decided to use two CPUs on the DB2 Connect system for further runs.

Scaling DB2 Connect MAXAGENTS

In these tests, we measured the affects of connection concentration at DB2 Connect. Two virtual CPUs were defined to the DB2 Connect z/VM guest and 1 GB of memory was used. The number of users was held constant at 500. The z/VM LPAR was assigned six CPUs and the z/OS LPAR was assigned four CPUs. The MAXAGENTS were varied from 50 to 500 and think time values of 2500 and 5000 milliseconds were used.

Figure 14 shows throughput for a DB2 Connect z/VM guest with think times of 2500 and 5000 and scaling MAXAGENTS from 50 to 500.


28

Figure 14. IRWW MAXAGENTS scaling throughput Figure 15 shows CPU utilization for DB2 Connect z/VM guest with think times of 2500 to 5000 milliseconds and scaling MAXAGENTS from 50 to 500.

Figure 15. IRWW workload - MAXAGENTS scaling Observations

Maximum throughput is reached at 100 MAXAGENTS in both the 2500 and 5000 milliseconds think time cases. More MAXAGENTS are not required for this


29

workload. In the 2500 milliseconds case there is a significant increase in throughput at 100 MAXAGENTS, compared to 50 users.

Conclusion

For the IRWW workload, a MAXAGENTS value of 100 is sufficient for this workload, even though the number of users is 500. This demonstrates that limiting the connections to DB2 on z/OS can be done without degrading of the throughput, up to a certain load level. It can also be used to limit the load created from this source on DB2 to a certain level, protecting the database against load peaks from that source.

Comparison of SQLJ versus JDBC

IRWW provides two methods for driving the workload. One uses SQLJ and the other JDBC. These tests compare the two methods. Two virtual CPUs were defined for the DB2 Connect z/VM guest and 1 GB memory was used. The think time values used for the IRWW workload driver were 2000 milliseconds and 5000 milliseconds and the number of user threads was held constant at 500. The z/VM LPAR was assigned six CPUs and the z/OS LPAR was assigned four CPUs. The parameter, MAXAGENTS, was set high enough that the number of active connections was not limited.

Figure 16 shows throughput for a DB2 Connect z/VM guest using SQLJ and JDBC with think times of 2000 and 5000 milliseconds.

Figure 16. IRWW workload transactions per second - JDBC versus SQLJ


30

Figure 17 shows CPU utilization for a DB2 Connect z/VM guest, z/VM LPAR, and z/OS LPAR using SQLJ and JDBC with think times of 2000 and 5000 milliseconds.

Figure 17. IRWW workload % CPU - JDBC versus SQLJ Observations

There is little difference between the IRWW JDBC and SQLJ workload drivers.

Conclusion

We selected the SQLJ workload driver because, in our environment, it produces slightly higher throughput at lower CPU utilization.

Compare DB2 Connect to non-DB2 Connect

In this test, the IRWW workload driver user sent all transaction requests directly to DB2 on z/OS, bypassing DB2 Connect. This test was run with 500 users and think times of 2000 and 2500 milliseconds. We also ran 100 users with a think time of 100 milliseconds. Two virtual CPUs were defined for the DB2 Connect z/VM guest and 1 GB of memory was used. The z/VM LPAR was assigned six CPUs and the z/OS LPAR was assigned four CPUs. The parameter, MAXAGENTS, was set high enough that the number of active connections was not limited.

Figure 18 compares the DB2 Connect runs to the non-DB2 Connect runs.


31

Figure 18. IRWW workload - DB2 Connect versus non-DB2 Connect

Observations

At a think time of 100 milliseconds and the lower number or users, bypassing DB2 Connect resulted in 30% higher throughput. As the think time and number of users increase, the difference in throughput decreases.

Conclusion

In the IRWW workload case, it appears that DB2 Connect has a higher impact on throughput than in the Trade 6 workload case, especially with a lower number of users. Bypassing DB2 Connect resulted in 6% to 30% higher throughput. The impact of DB2 Connect is higher – the shorter the think time. This is expected because here the request rate is higher and DB2 Connect has more activity to manage.

Appendix A. Detailed set up examples

This appendix contains detailed examples of configurations and sample scripts we used in our test runs.

DB2 Connect configuration commands for Trade 6

The following commands were used to define the Trade 6 database on the DB2 Connect guest.

db2set DB2COMM=tcpip db2 catalog tcpip node trade6db remote 10.10.50.110 server 446


32

db2 catalog dcs database trade6db as DB91ZOS db2 catalog database trade6db as trade6db at node trade6db authentication DCS

You can issue the following DB2 command to generate the output shown below.

db2 connect to trade6db user IBMUSER using password

Database Connection Information Database server = DB2 OS/390 9.1.5 SQL authorization ID = IBMUSER Local database alias = TRADE6DB

You can issue the following DB2 command to generate the output shown below.

db2 bind /home/db2inst2/sqllib/bnd/@ddcsmvs.lst grant public blocking all V sqlerror continue

Note:

The V symbol indicates that the text continues on the next line. These lines should be entered on one line, not broken into multiple lines.

LINE MESSAGES FOR ddcsmvs.lst ------ -------------------------------------------------------------------- SQL0061W The binder is in progress. LINE MESSAGES FOR db2clist.bnd ------ -------------------------------------------------------------------- SQL0038W The bind option SQLERROR CONTINUE has been activated since it is required when binding this DB2-supplied list file to DB2/MVS, SQL/DS, or OS/400. SQL0038W The bind option SQLERROR CONTINUE has been activated since it is required when binding this DB2-supplied list file to DB2/MVS, SQL/DS, or OS/400. LINE MESSAGES FOR db2clpcs.bnd ------ -------------------------------------------------------------------- 3521 SQL0199N The use of the reserved word "UNION" following "" is not valid. Expected tokens may include: ", )". SQLSTATE=42601


33

LINE MESSAGES FOR db2clprr.bnd ------ -------------------------------------------------------------------- 3521 SQL0199N The use of the reserved word "UNION" following "" is not valid. Expected tokens may include: ", )". SQLSTATE=42601 LINE MESSAGES FOR db2clpur.bnd ------ -------------------------------------------------------------------- 3521 SQL0199N The use of the reserved word "UNION" following "" is not valid. Expected tokens may include: ", )". SQLSTATE=42601 LINE MESSAGES FOR db2clprs.bnd ------ -------------------------------------------------------------------- 3521 SQL0199N The use of the reserved word "UNION" following "" is not valid. Expected tokens may include: ", )". SQLSTATE=42601 LINE MESSAGES FOR ddcsmvs.lst ------ -------------------------------------------------------------------- SQL0091N Binding was ended with "0" errors and "6" warnings.

Buffer pools used with the Trade 6 database on z/OS

Table 5. Buffer pools used with Trade 6 database on z/OS

Buffer Pool ID Size 9 60000

10 60000

DB2 Connect configuration commands for IRWW

db2 catalog tcpip node irwwdb remote 10.10.50.110 server 446 db2 catalog dcs database irwwdb as DB91ZOS db2 catalog database irwwdb as irwwdb at node irwwdb authentication dcs


34

Buffer pools used with the IRWW database on z/OS

Table 6. Buffer pools used with IRWW database on z/OS

Buffer Pool ID Size 1 60000

2 60000

3 10000

4 7000

5 10000

6 20000

7 20000

8 60000

Appendix B. Other Sources of Information

For information on WebSphere Application Server see:

http://www.ibm.com/software/info1/websphere/index.jsp?tab=products/

apptransaction

For information on Linux on System z see:

www.ibm.com/servers/eserver/zseries/os/linux/

For information on z/VM see:

www.vm.ibm.com

For information on IBM open source projects see:

www.ibm.com/developerworks/opensource/index.html

For information on DB2 9 see:

http://www.ibm.com/software/data/db2/udb/support/manualsv9.html

For information on DB2 Connect see:

– Quick Beginnings for DB2 Connect, GC10-4243 at:

ftp://ftp.software.ibm.com/ps/products/db2/info/vr9/pdf/letter/en_

US/db2c6e90.pdf

– DB2 Connect User's Guide, SC10-4249 at:

ftp://ftp.software.ibm.com/ps/products/db2/info/vr9/pdf/letter/en_

US/db2c0e90.pdf

– An introduction to IBM DB2 Connect: It's more than meets the eye:

The basics of DB2 Connect at

http://www.ibm.com/developerworks/db2/library/techarticle/dm-

0503katsnelson/index.html

For information on Workload Management see:

– z/OS MVS Planning: Workload Management

For information on DB2 for z/OS see:

– DB2 for z/OS Performance Monitoring and Tuning Guide


35

© Copyright IBM Corporation 2008

IBM Corporation New Orchard Rd. Armonk, NY 10504 U.S.A.

Produced in the United States of America 2/08 All Rights Reserved

IBM, IBM logo, DB2, DB2 Connect, Enterprise Storage Server, HiperSockets, MVS, System x, System z, System z9, WebSphere, z/OS, z/VM and zSeries are trademarks or registered trademarks of International Business Machines Corporation of the United States, other countries or both.

The following are trademarks or registered trademarks of other companies

Java and all Java-based trademarks and logos are trademarks of Sun Microsystems, Inc. in the United States, other countries or both.

UNIX is a registered trademark of The Open Group in the United States and other countries.

Intel, Intel logo, Intel Inside, Intel Inside logo, Intel Centrino, Intel Centrino logo, Celeron, Intel Xeon, Intel SpeedStep, Itanium, and Pentium are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries.

Linux is a registered trademark of Linus Torvalds in the United States and other countries.

SUSE is a registered trademark of Novell, Inc., in the United States and other countries.

Other company, product and service names may be trademarks or service marks of others.

Information concerning non-IBM products was obtained from the suppliers of their products or their published announcements. Questions on the capabilities of the non-IBM products should be addressed with the suppliers.

IBM hardware products are manufactured from new parts, or new and serviceable used parts. Regardless, our warranty terms apply.

IBM may not offer the products, services or features discussed in this document in other countries, and the information may be subject to change without notice. Consult your local IBM business contact for information on the product or services available in your area.

THE INFORMATION CONTAINED IN THIS DOCUMENTATION IS PROVIDED FOR INFORMATIONAL PURPOSES ONLY. WHILE EFFORTS WERE MADE TO VERIFY THE COMPLETENESS AND ACCURACY OF THE INFORMATION CONTAINED IN THIS DOCUMENTATION, IT IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED. IN ADDITION, THIS INFORMATION IS BASED ON IBM'S CURRENT PRODUCT PLANS AND STRATEGY, WHICH ARE SUBJECT TO CHANGE BY IBM WITHOUT NOTICE. IBM SHALL NOT BE RESPONSIBLE FOR ANY DAMAGES ARISING OUT OF THE USE OF, OR OTHERWISE RELATED TO, THIS DOCUMENTATION OR ANY OTHER DOCUMENTATION. NOTHING CONTAINED IN THIS DOCUMENTATION IS INTENDED TO, NOR SHALL HAVE THE EFFECT OF, CREATING ANY WARRANTIES OR REPRESENTATIONS FROM IBM (OR ITS SUPPLIERS OR LICENSORS), OR ALTERING THE TERMS AND CONDITIONS OF THE APPLICABLE LICENSE AGREEMENT GOVERNING THE USE OF IBM SOFTWARE

ZSW03032-USEN-00

Performance of environments using DB2 Connect Enterprise ...public.dhe.ibm.com/software/dw/linux390/perf/ZSW03032-USEN-00.pdfThe objectives of this project were to analyze the performance

Documents