No one can argue the benefits of implementing a data warehouse. A data warehouse can provide additional insight about your customers, processes, and vendors. Through the use of data mining, you can determine hidden predictive information from your database. This new information allows you to better address the needs of your business and have an edge on your competition.
But in order to have a data warehouse, you may have to acquire a large system. The technical specifications for a data warehouse are great. A highly tuned batch system is required to perform all the necessary functions of a data warehouse. In the past, you needed a separate system with a legion of disks to remove the workload from the online transaction processing (OLTP) system. Purchasing this type of server with all the necessary components is hard to justify—the costs can be great.
But what if you could combine a new data warehousing system with an upgrade to your OLTP system or combine your data warehouse and datamarts? What if you could have both an OLTP system and a data warehouse residing on the same system with each tuned for its unique processes? This is now possible with the release of OS/400 V4R4.
Logical Partitioning (LPAR) allows for several AS/400 servers to run independently of each other on the same system. LPAR is achieved by distributing the resources on a single system to allow for the simultaneous running of multiple servers within a single symmetric multiprocessing (SMP) AS/400e, SXX, 6XX, or 7XX system. Each partition is assigned its own processors, disk, and memory. Logical Partitioning is automatically available after installing OS/400 V4R4; there is no additional charge.
Types of LPAR
There are two different types of Logical Partitioning: bus- and input/output processor (IOP)-level partitioning. Each type provides advantages depending upon the requirements.
Bus-level partitioning provides isolation on devices attached to the bus. This allows the operating system to have no dependencies on other partitions, which makes the management of the IOP resources simpler, as partitions cannot share the resources. With bus-level partitioning, you will get better performance, as the Systems Licensed Internal Code (SLIC) does not have to call the Partition Licensed Internal Code (PLIC) to queue commands for the bus. When dealing with bus-level partitioning, you do not have to deal
with the difficult dedicated service tools (DST) or system service tools (SST) to configure the devices.
IOP-level partitioning allows for sharing of IOP resources when the device is not in use by another partition. When the hardware can be shared, it will result in lower hardware costs. Management of the hardware is done through DST or SST. This requires personnel with DST or SST experience to perform these tasks. As a precaution, incorrect use of the DST or SST services may result in damage to the data. If there are any problems at the bus level, they can potentially cause problems on multiple partitions, resulting in no partition accessing the hardware.
Which type of partitioning is better? It really depends on your requirements. If there is a large demand for a piece of hardware on a particular partition, bus-level partitioning may be the better choice, as the hardware is dedicated to the partition and it does not have to fight with other partitions for rights to the hardware. But if the hardware is not a huge requirement and sharing is possible, then IOP-level partitioning is more appropriate.
LPAR is configured through the primary partition, which is the first operating system on the box. It is responsible for allocating the resources to the other partitions. Partition creation and management as well as resource allocation are accomplished with the options in DST or SST.
Implementing a logical partition requires some planning to ensure that sufficient resources are assigned to each partition. You must ensure that the systems have enough resources to run at their proper levels. To create a secondary partition, you must assign the hardware resources of the partition through the DST options. This assignment is done after they have been “de-assigned” from the primary partition. By default, all resources are assigned to the primary partition. After you have selected the resources, you will be prompted to select the load source disk unit, console device, alternate IPL device, and optional ECS adapter. As soon as that is done, you can begin installing the licensed internal code, OS/400, licensed programs, PTFs, and application software. The installation of OS/400 on the secondary partition can be done while the primary is operating. It will not affect the operation of the primary partition.
To configure LPAR on your AS/400, you must have an AS/400e, SXX, 6XX, or 7XX system. LPAR cannot be configured on a 170 because it requires a separate disk IOP for each partition. The primary partition requires a minimum of 256 MB of main storage, and the secondary partitions require a minimum of 64 MB of main storage. The number of partitions is limited to the number of processors on the system. Each partition must have a minimum of one processor.
Figure 1 shows one possible LPAR configuration for an AS/400 with 10 processors.
When implementing an LPAR solution, you may run into a couple of problems. For example, if a primary partition fails, the remaining partitions will fail as well. LPARs are controlled through a layer of the kernel code known as the hypervisor. The hypervisor manages the resources for all of the logical partitions that are defined on the system. The hypervisor layer is located in the code in the primary partition. Because of the interconnection to the primary partition, any resulting failure will cascade to all of the partitions on the system. So precautions like a UPS will help keep the primary partition continually active. In the event of a power failure, with the assistance of a UPS, the primary partition will power down all of the secondary partitions before it brings itself down.
After your users begin to realize the value of the data warehouse, they will become increasingly dependent upon it. So the availability of the information in the data warehouse
be-comes a greater issue, as it needs to have quick performance. With LPAR, each partition can be configured to handle a different type of workload. Your OLTP system will be tuned for interactive processing, and the data warehouse will be tuned for client/server processing. The amount of workload the partition can use is determined in the planning stages and is configured when the partition is created.
Advantages of LPAR
A major advantage in implementing an LPAR solution is the ability to have different versions of OS/400 running on one system. Some applications are incompatible with the newer versions of OS/400 and require a different version. This allows you to keep your applications on the previous versions of OS/400 and still consolidate the servers in the LPAR solution.
Consolidation of these servers can greatly affect IS costs. Shared hardware will result in more experienced personnel overseeing the resource management to ensure that it is done correctly. With less hardware to manage, there will be fewer hassles and problems, right? No, not really. Each partition should be considered a separate system and should be considered the same as managing separate servers. Each server will have its own unique aspects and will still require the same amount of maintenance to ensure proper execution. What LPAR does provide is the simplicity of server consolidation.
If you are planning or currently have a physical/logical three-tier data warehouse, you will have noticed that there are multiple AS/400s involved. You will need one for your OLTP system, another for the data warehouse, and one or more for your datamarts. The maintenance for all of these systems can become quite cumbersome, especially if they are not in the same location. The LPAR solution would allow you to consolidate all of these servers into the same system with each configured for its own workload type. This results in fewer physical devices to manage.
Enhancements in the V4R4 release of Management Central will aid in the implementation of an LPAR solution, as it makes the management of all the partitions a simpler task. It allows administrators to package AS/400 objects to be distributed amongst the other partitions or other AS/400 servers. This can be used to copy portions or all of a database for replication, cleansing, and transformation testing. It can also be used for full or partial system redundancy for backup purposes.
Management Central has many enhancements that assist in the management of PTF levels on your AS/400 partitions. PTF levels can now be monitored, compared, distributed, and installed across multiple partitions. This allows the PTFs to be installed from a system where they are proven to work. With PTF management, you will ensure that all servers are up-to-date with their PTFs. This is possible only if the systems are on the same version of OS/400.
Hardware and software inventory of the partitions can now be viewed, searched, or exported to a spreadsheet using Management Central. This inventory can help in determining the current owner of the hardware, which is especially helpful when IOP resources are being shared between partitions.
Management Central has the ability to run commands on multiple partitions and examine the results using the graphical interface. This can be helpful when the replication process needs to be immediately initiated. It can also be used to view the status and perform administrative tasks involving the database.
Management Central offers the ability to control the performance data that is collected on the partitions. When database performance is poor, the data can be collected and analyzed to determine the cause. These collections can then be distributed to other systems for analysis using performance tools. This is especially useful with Management Central’s ability to distribute the objects amongst other servers.
Licensing of the software becomes a great advantage for LPAR and data warehousing. AS/400 licensing is based on the serial number of the physical box, not the
individual operating system. Prior to LPAR, you would have been required to purchase software for each individual system. When dealing with a partitioned system, you will not be required to purchase another copy of the operating system for each partition unless V4R4 is needed. This is also true for many software vendor packages. Because the OLTP
system and the data warehouse are in different partitions, you will need only one version of the replication software. Proper execution still requires the software to be installed and configured on each partition.
Probably the most important aspect in LPAR that would be a major advantage to data warehousing is the capability of faster replication. Communication between partitions can be done in two different ways, as shown in Figure 2. The first method is external, through the use of standard LAN/WAN communication devices, wherein each partition would require its own LAN adapter installed and configured with the appropriate communication protocol. This method does not provide any benefits for data warehousing, as it is no different than what is currently available through normal communication methods.
The second method uses a version of OptiConnect that is called Virtual OptiConnect. This method provides faster communications between the partitions compared to the LAN/WAN communication devices. Virtual OptiConnect can be installed and configured for high-performance interpartition communications and requires only a single license of OptiConnect for the system.
This method of communication uses the bus bandwidth to communicate with the partitions. Virtual OptiConnect uses direct memory access (DMA) to transfer data from the disk of the source partition into memory. As Figure 3 illustrates, Virtual OptiConnect then copies the contents in memory from the source to the memory of the target partition. Once it has finished transferring the information to the target, it copies the information to the
target’s disk. Its performance is incomparable to external LAN/WAN devices, as the data is not required to travel through the network or involve communication IOP.
This method of communication can be beneficial to data warehousing. If replication, cleansing, and transformation can be completed quicker, the processes involved will not require as much processing power and, in turn, will result in more CPU available for other processes, not to mention that it will cut down on network traffic and free up hardware resources for other processes.
A Simple Plan
The time and cost of getting a data warehouse operational are immense. It requires a great deal of commitment to the project and to the hardware. LPAR will simplify some of the aspects regarding your data warehousing solution. It will reduce some of the cost involved in the acquisition of the components necessary for your solution. Sharable resources will help in the configuration.
There are many benefits to having LPAR in your data warehousing solution. It will greatly reduce the cost of software and hardware depending upon the LPAR type. It will provide faster communication between partitions, which will greatly reduce the amount of time necessary to replicate the data. All of these capabilities will greatly increase the performance of a data warehousing solution on an LPAR system.
4-way Interactive 4-way Client/Server
Figure 1: Using LPAR, this AS/400 system of 10 processors is divided into three servers.
4-way Interactive 4-way Client/Server
Figure 2: Partitions may be connected via a LAN adapter or Virtual OptiConnect.
Figure 3: Under Virtual OptiConnect, data is transferred through memory.