Optimizing IT Performance

Optimizing IT Performance End user based Performance Management Solutions bridge the gap to the business Time and the actual usage of conventional data center or server-based performance management solutions have clearly demonstrated that these types of products are unable to provide the critical data needed to effectively manage the performance of an organization s IT infrastructure. The deficiencies of these types of solutions extend from an inability to monitor end user client/server as well as web-based services to the requirement of having to configure or model the solution to encompass the endless number of variables that must be considered. The net effect of these deficiencies and other issues is the subject of this white paper. This document will attempt to comprehensively cover the subject matter and provide the reader with sufficient information for the evaluation of any performance management solution. Introduction Over the last few years IT departments have been under increased pressure to optimise resource utilization and to become more responsive to changes required by the business. The pressure to be more responsive and reduce costs can be especially daunting for IT organizations with complex installations, e.g. geographically distributed or transaction heavy applications. The demand to manage these complex information infrastructures with fewer resources is forcing IT professionals to reevaluate the methods and tools currently being used to accomplish these tasks. Conventional data center or server based solutions cover many aspects of an organizations infrastructure such as the servers and the network. However, the vast majority of time is not spent on managing devices but rather on dealing Features Award winning software. Used by IT professionals in 67 countries. One day to install. Company wide endto-end Application Performance Monitoring from day one. 18 built-in definitions for standard client/server applications for fast application performance monitoring implementation. Auto-detects standard applications. Easy setup with a few clicks in the Configuration Wizard. More than 200 standard reports with analytics for desktop, servers, network and citrix systems. Custom dashboard views for any mangament context (technology teams, help desk, CIO etc.). Auto-detects new components. Auto-suggests baselines/thres-holds. Speaks ITIL. Availability. Capacity and Change Management documentation and metrics for continual service improvements. Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 1 of 8

End user-based end-to-end performance management gives the IT executive management team the ability to very accurately report to the business on how the service is rendered as well as the planning information essential to accurately focus and document where investments are required with issues associated with cross-platform service level monitoring, performance optimization, capacity planning, identifying and resolving end user problems. Unfortunately, most conventional performance tools do not provide an end-to-end solution that effectively addresses the end user perspective. As an example, it is very difficult for a traditional three-tier infrastructure to detect why an end user in a remote office is experiencing performance problems only during parts of the day. Such periodic or non-standard problems are the most difficult, time consuming and expensive to overcome. Most enterprise management solutions are based on configuring a system for device and server alerts. These complex issues are extremely difficult to solve using any management architecture that requires configuration or modeled simulations to identify a measurement point. Since there are millions of alternative cross-referencing scenarios to consider, it becomes literally impossible to configure or model every possible combination of measurement points. The inability to consider all of the variables seriously reduces the IT manager s ability to identify or resolve periodic problems that arise. Independent research has concluded that almost 90% of support resources are consumed exclusively on problem identification. The first step to resolution, suggested by Forrester, is to acquire or to augment an existing solution with an end user-based performance management product. The new end user tools are able to measure the health of the entire IT infrastructure. This includes servers, networks, end user PCs, devices and applications, both client/server and web based services. The collected data from the end user based tool will provide the operational monitoring with the intelligence needed to resolve performance, capacity and configuration constraints as well as cross system error resolution. This paper will examine the elements involved in gaining an actual view of end user application performance to quickly detect IT system issues, and how Performance Guard from CapaSystems can help companies proactively manage their IT infrastructure. Systems Management Challenges There are two main issues facing IT and Service Desk managers today. 1. Obtaining a holistic view of the enterprise IT performance by quickly, often in real time, acquiring the data needed to accurately identify and solve IT problems. 2. The second issue is the ability to leverage this information to proactively solve problems and optimise capacity before the end user is adversely impacted. It is an established fact that monitoring and measuring application performance at end user level will help ensure IT system availability. Conventional data centerbased solutions fall short of delivering all of the information needed to identify and resolve cross platform system issues quickly. This lack of information generally stems from the fact that configuration options are missing or the available tools are based on simulations. Simulations cannot provide any actual end user data nor are they able to detect period degradation in response times or trending. This makes it very difficult to trace periodic latencies and identify and plan for capacity problems. When a system performance problem arises the IT manager typically needs to pursue a time consuming, expensive and tedious process to identify and resolve any given issue. The delay and downtime caused by this time consuming process is a source of frustration for IT management at every level. An even more important consideration is the costs associated with this process. IT Managers need the ability to quickly and accurately identify problems. They need coherent performance data on applications, networks and system components that allow them to proactively manage IT systems. The availability of this performance data serves several functions. First, it allows IT Management to optimise application service, quality and availability. Second, it provides the essential planning intelligence required to identify the specific components needed to upgrade or enhance the infrastructure as well as the documentation to justify these expenditures. Thirdly, access to this data results in rapid problem resolution and greater end user satisfaction. Most importantly it will dramatically lower costs. Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 2 of 8

User based performance management Most data center solutions manage server resources very well. To achieve end-to-end management the IT professional needs to evaluate system performance based on the end user experience. The end user is the IT customer and all services from application availability to performance problems must be measured and evaluated from the customer s perspective. Data center or server based solutions do not capture this end user data or perspective. End user solutions utilize a monitoring agent installed on the end user device. This agent measures all local, network, server and application activity and nonintrusively reports this to a central IT performance database server. The data provided is a real time view of the end user experience. IT management can use this intelligence for a multitude of tasks such as resource planning and measuring service levels. All levels of the Service Desk support staff can use it to quickly identify problem areas and to proactively detect a degradation trend in the delivery of IT services. Passive agents also eliminate the introduction of any additional load on the existing IT infrastructure which is typically experienced with alternative solutions. The endless variables associated with a solution that must be configured or modelled are automatically captured by an end user based system. This is accomplished by initiating the capture of information at the end user and then looking back across the network. Regardless of the size or complexity of the network, this method of capturing performance data encompasses the entire IT infrastructure, including web based services and applications. PerformanceGuard provides drill down resolution that allows for before and after scenarios to determine what periodic problems actually consisted of, how they should be solved and the impact With PerformanceGuard we often identify potential performance issue so fast that we fix the issue before the end-users have even reported it. We can track tendencies through our measurements and act proactively before problems become a ticket in our sevice desk. DSB Informatics A/S of the proposed solution. This data will ensure that the investments in the IT infrastructure deliver the required performance based on set baselines. The PerformanceGuard product from CapaSystems is an award winning product that is the first end-to-end performance management solution commercially available. PerformanceGuard delivers a complete application, server and network monitoring solution, which includes information on PC resources. Performance- Guard provides exact measurements of the end user experience including an up to the minute overview of all components and applications touched by any end user. The PerformanceGuard Solution The PerformanceGuard architecture consists of an agent on each PC and a collection/repository/reporting server. PerformanceGuard agent The PerformanceGuard agent is installed on each PC using standard software distribution methods. It passively monitors and collects real end user performance measurements using a lightweight agent installed on individual machines. The agent collects response time and other performance metrics on the end user PC. Agent Frontend Server Database Alarm monitor Database Database Backend Server Alarm monitor Agent Frontend Server Alarm monitor Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 3 of 8

It also looks back across the network from the PC and collects performance data on the entire IT infrastructure, including networks, network components like servers, bridges and routers and all applications, both client/server and web based. Performance- Guard collects the data into local reports and at secified time intervals sends the data to the PerformanceGuard Server. PerformanceGuard Server The PerformanceGuard Server receives, stores, aggregates and displays all of the vital statistics and performance measurements that are collected by the agents. This server consists of the following subcomponents: The Frontend servers (optional) collect data from the agents and store information on end user level in the Front end database. This data is aggregated into subnet level and is then sent to the back end server. communication between the agents and the Frontend server is performed using TCP/IP. The Backend server collects data form one or more Frontend servers and inserts the data in the Performance Guard Data Repository (database). The Data Repository is a standard database server based on Microsoft SQL Server. All connections to the database use the open JDBC interface. The Display accesses data from the Performance Guard database through a user friendly display containing a large set of predefined reports. The data can also be configured in a multitude of user defined formats such as data tables, graphs and configuration diagrams. PerformanceGuard allows the IT manager to either analyse historical data or monitor data in real time. Performance- Guard is compatible with a number of reporting tools, among others Crystal Reports. Simple Installation and Setup The central PerformanceGuard server is installed via a wizard interface. Agents are distributed to all clients using any standard software distribution tool or a link to an install file. The agents install automatically and begin reporting to the PerformanceGuard server immediately after installation. In standard installations the agent will consume less than 1% of the end user devices CPU and memory resources. All maintenance of the agent is centralized and performed automatically when the agents report data to the server. Fast Problem Detection In order to appreciate the response times reported by PerformanceGuard it is important to understand the fundamentals of how PerformanceGuard measures response time. Response time is measured at the TCP/UDP IP level. TCP/IP is the most commonly used protocol today. Only packets that contain application data are considered when PerformanceGuard measures response times. For the first time ever, IT can get its own Executive Dashboards that show the end-toend performance of all systems and networks over time or in real-time, as experienced by the users. Information on critical service degradation with advanced drill down detailed information is only a click away. Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 4 of 8

A REAL LIFE EXAMPLE Example: Resolve SAP performance issues Identifying a specific performance issue for an application - such as SAP delivered via Citrix over a WAN - is extremely difficult due to the large number of variables. often, all silos will report that the systems are fine. Yet the end user experiences poor performance. en duser performance management is the only way to quickly identify the performance bottleneck and resolve it. Illustration 1 show performance degradation from a set of SAP users in New York experiencing poor performance on the SAP system in Germany during afternoon peak hours. Solution It is the combination between the two delays that made SAP grind to a halt. However, the default soultion adding more expensive SAP servers or capacity over the Atlantic would be very expensive and still not solve the issue. Better local network capacity in New York will reduce the response time to an acceptable level and solve the problem at a much lower cost. Unexpectedly, the first performancehot-spot is response between the end-users network segment and the internet. The end-to-end monitoring diagram shows two performance hot-spots The second red problem area is the transmission time between New York and Europe. The delay is due to the distance, not capacity and consequently cannot be solved by more bandwith. When a client sends a request to a server it sends one or more packets. The server then processes the request and sends one or more packets back to the client. The measured response time is the elapsed time between the last sent client request packet and receipt of the first reply packet from the server. An agent aggregates response time measurements based on the server IP address and TCP/UDP port on which the server and client communicate. For example, response times for all communications with a specific Web server within a single report period will include the following information (which will be sent to the Frontend Server). Response times for all communications with a specific web server include the following information: Accumulated response time Number of connections Number of trains sent and received (The term train is used when multiple packets are sent in one direction before any packets are sent in the opposite directions). Number of bytes sent and received The response time for the combination of agent server service is then calculated by the back end as an accumulated response time divided by the number of received trains. In order to display response time from measurements taken on multiple clients it is necessary to aggregate the data further. Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 5 of 8

In this case the response time of a group of agents and a specific server service is calculated as the sum of accumulated response times divided by the sum of received trains for all agents within the group. These metrics become the foundation of all aggregated data which is utilized to deliver the corresponding graphs, comparisons, and other related measurements. The information is used by IT managers for rapid error detection, proactive problem solving on the component level, and ongoing system performance tuning. Fast Isolation of Local Resource Issues and Problematic Processes The PerformanceGuard Agent captures the following local (pc) metrics: Resource Metrics CPU Usage: Percentage of CPU time not spent running idle. Free Physical Memory: Amount of physical memory available for allocation. Free Paging File: Amount of paging file space available for allocation. Virtual Memory: Amount of virtual memory available for allocation. The Performance Guard Agent captures the following local process metrics: CPU Usage: Percentage of available CPU time used for a certain process. Memory Usage: Number of bytes that a process has allocated that cannot be shared with other processes. Thread Count: Number of operating system threads used by a process. Handle Count: Number of operating system (Windows) handles used by a process. Values for these metrics are sampled at regular (user adjustable) intervals. The maximum and average value is reported for each sampled interval timeframe. Windows has many idle processes running that are of no interest; therefore a limit may be set on the number of processes monitored by PerformanceGuard. These limits can be imposed at the CPU, memory, I/O level or within the exact process to be monitored. Component Groupings are the key to Problem solution The most important requirement in setting up the Performance Guard system is in the grouping of agents. Data aggregation is performed at the group level. Agents can be grouped in several different ways. An agent can belong to multiple groups that represent a location, job function and / or application usage. By default an agent is included in a dynamic network group containing all agents within a subnet. If an agent moves to another subnet, it will automatically change network group in PerformanceGuard. Once an agent has been installed, PerformanceGuard will automatically detect all servers and services that the agents are accessing. To assist in the process of defining which servers and services to monitor, PerformanceGuard will list all accessed servers and services ranked by activity level. Once the known servers Figure 3. Servers and services ranked by activity level Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 6 of 8

and services have been selected, the required groupings can be established. Like agents, both servers and services can be assigned to more than one group. As an example, viewing the average response times for a set of servers and agent groups, in a given time period and for a given service, provides visibility into whether some groups of agents have better response time than others. If the groups of agents are geographically separated, this could indicate that there is a network problem with one of the groups. Realtime Data Output for High Visibility Once the groupings have been defined, reports generated and input entered into the repository, the information can be extracted and used to optimise IT systems. Data can be generated in any combination of agent, server or service groupings for a given time period and displayed using graphs that reflect information, such as: Response Time Accumulated Histogram (%) Requests per Second Active Agents Sent Bytes per Second Sent Packets per Second Received Bytes per Second Connections per Second Connections Reset (%) By comparing all agents and response times against two applications it becomes obvious where IT managers need to concentrate their efforts to correct the problem. Collected Data for RealTime or Historical Reporting Custom reports are a collection of graphs that provide IT managers with an overview of the service delivered by either a specific application or a number of applications. The IT manager may customize graph formats and the appearance of graphs to be added to any report. The reports can be accessed via a standard web browser or sent by email as a PDF attachment on a regular schedule. Application Ping Feature Even though PerformanceGuard is primarily used as a passive monitoring tool, there are situations, such as the need to measure application availability, where an active agent may be beneficial. By using the application ping feature it is possible to dictate when and how often an agent or group of agents actively pings a server for availability and response times. There are three types of active pings to choose from: 1. Standard ping: used for testing the server 2. HTTP ping: used to test if a web server present 3. Trace ping: This unique feature enables the user to track the response from every single node in the network. The ability of the agent to act either passively or actively, as required by the IT manager, is unique to third generation. Trace Ping is Critical For virtually every IT organization understanding application, server and network performance is critical. Unfortunately, networks often represent a mix of lines that are a virtual black hole to most IT managers. The Trace ping feature eliminates network black holes by allowing the user to locate network issues. The Trace ping identifies problems down to the individual node level in a carrier network, without the need to reference logs or outside data. In the prior SAP example, the trace ping feature easily identified the path to the right resolution. Without an end user based management solution an organization can spend substantial sums of money on solutions that are completely ineffective. The SAP example is very common and in such circumstances an end userbased management solution can pay for itself in days or weeks. Syscom Name Period Availability Response time (ms) Response time (%) Time Passive items Active items Target Actual Target Actual Target Actual Day 95% 100% 100 74,87 90% 96,26% 12-09-11 CEST Day 95% 100% 100 194,49 90% 76,47% 13-09-11 CEST Day 95% 100% 100 231,44 90% 67,1% 14-09-11 CEST Day 95% 100% 100 73 90% 96,24% 15-09-11 CEST Day 95% 100% 100 96,46 90% 99,00% 16-09-11 CEST Day 95% 100% 100 74,45 90% 92,88% 17-09-11 CEST Day 95% 100% 100 81,6 90% 87,11% 18-09-11 CEST Transaction items Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 7 of 8

Service Level Agreement A Service Level Agreement (SLA) report is a collection of items such as minimum response time criteria. An SLA item can be either an active item, such as a ping from the Performance Guard agent against a predefined application, or a passive item that is a normal passive response time measurement. The SLA item contains threshold values for response time and availability. If these thresholds are exceeded the SLA item is considered to have failed. Alarms Sound Alarms can be configured and defined as a point in time where the associated baseline alarm threshold has been exceeded. The alarm itself is defined on the agent server service grouping and is sampled once every minute by the backend database. If the threshold is exceeded, the alarm is triggered and its status is displayed on the PerformanceGuard dashboard. The alarm can also be sent as an email or an SNMP trap for showing in external monitoring tools such as IBM Tivoli, Microsoft SCOM etc. End user alarms allow the IT manager to rapidly identify the problem source and proactively take the necessary corrective action. Advanced enviroments such as Citrix Many large organizations take advantage of thin clients such as Citrix. PerformanceGuard contains unique modules that specifically address the resource utilization of both the PCs having the Citrix ICA client installed and the Citrix Server farm. Identifying the source of performance issues with Citrix installations can be problematic. End user performance management solutions instantly measure the performance at each level of the installation and pinpoint the areas of constraint whether it is the source application, Citrix server, network or the end user client. Without an end user based solution, such as Performance Guard, these scenarios often present the IT professional without any effective insight into the complete deployment. Enterprise Management Framework PerformanceGuard can be used as a standalone solution or to enhance and complement any traditional data center solution. When used to enhance conventional data center products integration at some level is frequently required. PerformanceGuard sends alerts with detailed information to enterprise management systems such as Microsoft Operations ManagerSCOM, IBM Tivoli, CA Unicenter or HP OpenView. This is accomplished using standards-based SNMP formats. The alerts contain detailed information and links to the appropriate reports. The ability to integrate these alerts allows the help desk staff to rapidly and proactively resolve any problems. The PerformanceGuard Advantage The PerformanceGuard approach to application performance monitoring is a unique one of a kind process. PerformanceGuard provides real-time measurements, alerts and reports that can instantly show trends and problems that have yet to be detected by standard infrastructure monitoring. This capability allows PerformanceGuard to provide the first CIO level dashboard that provides an end-to-end view of IT performance. PerformanceGuard supplements infrastructure management tools and shows the health of an application from the end user perspective while delivering a true productivity view of the services received by the end users. This capability allows the IT manager to optimise system performance, increase productivity, improve end user services and substantially reduce expenses on their infrastructure. About CapaSystems Software Solutions for Device Management, Performance Monitoring and end-user satisfaction: At CapaSystems we are dedicated to creating a future where smart technologies will eliminate all manual processes. We do this by providing expertise and smart, proprietary software that unleashes the full potential of our customers IT-operation. CapaSystems is a Danish software vendor established in March 1996. Today we have customers worldwide and strong partner channels in Scandinavia, Central & Eastern Europe, the USA and the Middle East. Roskildevej 342C, DK-2630 Taastrup, Denmark. Tel. +45 70 10 70 55. www.capasystems.com 8 of 8