Description: By providing unlimited capacity, continuous availability and application transparency, DB2 pureScale reduce the risk and cost growth. DB2 pureScale on IBM Power Systems combines PowerHA pureScale technology, on a UNIX or x86 systems deliver unparalleled levels of database scalability and availability. The product is the undisputed overall system availability, scalability, security and reliability of leading product DB2 for z / OS and System z to a powerful supplement.
In the economic recovery process, with instant access to core business data is always the survival of enterprises, as well as critical success factors. As more and more dollars into the domestic market, companies need help with high availability and suitability of infrastructure to improve their agility to be able to seize new opportunities.
Most distributed software companies will be available in the market when the level of "mainframe class" or "5-9" linked to the availability of such terms. These phrases are trying to convey the industry-recognized high-availability "gold" standard (that is, DB2 for z / OS) the continued availability of the target set.
|Availability of downtime per year|
|99.9%||8 hours 20 minutes|
|99%||3 days 11 hours 18 minutes|
|95%||18 days 6 hours|
|90%||34 days 17 hours 17 minutes|
|85%||54 days 18 hours|
Today, the availability not only means being able to calmly deal with component failure and to resume normal transaction processing. If your service level agreement (SLA) specifies the expected query response time should be in a few seconds, the server has to spend one minute before returning the query, then this is the usability issues. To ensure availability, your system not only provides transaction services, but also within the period specified in the SLA service.
For example, if the business cycle, causing seasonal fluctuations in the availability of extensions to the problem, the availability of truly transparent framework must be able to increase the resources, while avoiding changes to the application to meet changing performance requirements. Transparency is a key factor: the increased capacity, should not be so aware of the application with the cluster (the application to know which node in which the data to avoid contention between nodes). Companies can not invest sufficient funds to develop these complex applications, it can not achieve a reasonable extension. Why is this? First, it is clear that the cluster-aware applications need to adapt the amount and distribution of data changing. Cluster-aware applications not only require code changes with the development of clusters: the code still needs to undergo testing, quality assurance (Q / A), deployment and certification process. This may cause companies to spend a few weeks to coordinate, and will inevitably run out of infrastructure should have a better use of resources.
Used in the distributed platforms (non-mainframe) on the expansion of database transactions and other products commonly used outdated architecture, so it will create unnecessary difficulties for the extension (such as overhead), making it impossible to ensure compliance with the SLA agreements.
IBM DB2 pureScale technology (hereinafter referred to as DB2 pureScale) can be real high availability and transparent application extensions combined in a system to meet your current and future continuous availability needs. IBM Power Systems servers and IBM storage solutions is the integration of DB2 pureScale architecture deliver this high-value solutions of the intrinsic base.
So far, the "mainframe class" is still a compelling marketing vocabulary. DB2 pureScale marks the first truly transparent and scalable architecture for distributed application platform. This article introduces the basic concepts of DB2 pureScale technology, background information, as well as its high availability and transparent application extensions with the unique advantages of mystery.
DB2 pureScale DB2 is a new optional feature that allows you to "double machine (active-active)" configuration database be extended to a set of servers to deliver high levels of availability and scalability. In this configuration, running on the host (or server) on the DB2 copy can be both read and write the same data.
Shared DB2 data server one or more DB2 data sharing group is called. DB2 data sharing group is the group server. Currently, the data sharing group to support the maximum number of members is 128.
Members, in addition to DB2, PowerHA pureScale ™ component also provides integrated management, and lock the global cache for data pages (called group buffer pools).
Data sharing group members can be a very effective InfiniBand ™ network components that interact directly with the PowerHA pureScale, as shown below. This means that members with centralized locking and cache a point to point between devices (P2P) connection.
You hear or see any description on mainframe-class availability, refer to the DB2 for z / OS to set the gold standard for usability. In fact, the world has no solution for a database can run on availability with DB2 for z / OS System z server par.
DB2 for z / OS data sharing, the underlying technology used to ensure that the server can continue to meet SLA requirements, as Coupling Facility provides a centralized lock management and global cache, which is quick to recover from a failure to provide a guarantee. In fact, DB2 for z / OS in the strict sense has achieved a "5-9s" availability level, while expanding the workload in a seamless linear enjoys a high reputation.
From DB2 for z / OS, many people will think of a wide range of scalability and high availability. This market reputation is not groundless, but from these systems in the database workload availability of market-leading position has no regret moving. Perhaps the best evidence of DB2 for z / OS than the power of Oracle founder and CEO Larry Ellison's comments:
Figure words: I made fun of too many other databases, but there was the mainframe version of DB2 have respect for the heart. It is well-deserved first-class technology.
DB2 for z / OS what is unique, so it let Ellison praised Canada? DB2 for z / OS data sharing in the field of "unique secrets" to its users too familiar, it is known Coupling Facility. Coupling Facility only for DB2 for z / OS gives a linear expansion of capacity, also provides a centralized device management locks. In addition, it also acts as dirty pages (dirty page) global shared buffer pool (helps scalability and recoverability of operation).
DB2 pureScale technology adhering to DB2 for z / OS Coupling Facility tradition of blood, so the accumulation of many advantages, so as to DB2 for z / OS availability and scalability to become the "gold" standard basis. This is how to do it? DB2 pureScale comes with an IBM powerHA pureScale component that provides the same centralized lock management and strict sense of the global shared buffer pool structure.
Other vendors have implemented the use of shared disk database architecture, one of the most influential is the Oracle Real Application Clusters (Oracle RAC). However, in the development and design of Oracle RAC, the distributed platform technology also allow for effective access to centralized shared cache. Result, Oracle RAC is designed ultimately to become a mock DB2 for z / OS a try; Oracle RAC is also distributed lock management and distributed caching architecture originated. The introduction of Oracle RAC shared disk architecture scale-even after the loss of DB2 for z / OS solution simplicity advantage. On the other hand, DB2 for z / OS and DB2 pureScale provide the same centralized resource management, therefore, to solve these complex problems of scalability and availability. This article will discuss this later in the content.
The most fundamental problem is that the market is only a framework to deliver a truly transparent application scalability and high availability. With modern hardware platform to achieve a distributed interconnect, and InfiniBand-based non-disruptive Remote Direct Memory Access (RDMA) in-depth development, DB2 for z / OS used focus lock and buffer caching algorithm is that it is no longer The exclusive patent. DB2 pureScale to the well-proven technology into the industry to a distributed platform, which also represents the progress of the IBM family.
In the horizontal expansion of the database environment, cost savings, the key is truly transparent application extension mechanism. Transparent extension means that the database engine for OLTP applications provide greater throughput and faster response times, and there is no requirement for data locality.
Local data required by the application of that data on a server in which it is connected, and contention between nodes is rarely the same data page. In the scale-out architectures, if the use of web-based messaging infrastructure to share data in a cluster, the local nature of the data is particularly important.
Rely on the local expansion of effective scale-out architectures requires developers to create complex business applications to achieve a cluster-aware nature. Cluster-aware application development and deployment of not only more complex and costs more expensive, but when the cluster changes Shihai requires redesigning the application. Some vendors may claim that their architecture can run any application without modification; but if not designed to achieve some form of cluster-aware nature, they will not extend any application.
Transparent application extension means that the application can not have cluster-aware of the use of scale-out architectures. DB2 pureScale is unique to the distributed platform, and its efficiency from the network and hardware of modern architecture, and pureScale centralized locking and caching mechanism to use.
In order to reduce the cluster communication between nodes in order to achieve lock management and global cache service, DB2 pureScale use powerHA pureScale cluster acceleration device (hereinafter referred to as CF) and RDMA technology to provide transparent application scalability.
RDMA allows individual members of the cluster in memory, direct access to CF, and CF members can directly access the memory. For example, assume that a certain cluster members (1 member) want to read is not stored in the local buffer pool data page. DB2 will be assigned an agent (or thread) to perform this transaction; then the broker uses RDMA write directly to the CF memory data, claimed that they needed to read a particular page. If members want to read page 1 in the overall concentration of the buffer pool of CF, the CF will be pushed directly to the members of a page in memory, rather than the members of the Acting Executive I / O operations read from the disk take it. Using RDMA, only members of a proxy server to initiate a remote memcopy (memory copying) call, thus avoiding the high cost of inter-process communication, processor interrupt, IP stack at the other. In short, pureScale allows members of the agency through the implementation of the local memory copy operation to perform remote memory copy operation.
These lightweight remote memory calls, along with centralized management of the buffer pool and lock the device, means that applications do not need to connect to that already contains data members. Any member of the cluster can effectively receive data from the global buffer pool page, no matter how much the cluster. Most RDMA calls very quickly, which makes the calling DB2 processes without waiting for the CF response to the allocated CPU time, and do not need to re-schedule to complete the task. For example, to notify the CF will update a row (and therefore need an X lock), a member of the agent to perform Set Lock State (SLS) request, the information is written directly into the lock on the CF memory. CF will confirm the other members of the cluster did not lock the line, and directly modify the request to approve members of the memory lock request. The SLS is only 15 microseconds to complete the whole process, so the agent does not need to give up the allocated CPU time. Agents can continue to run efficiently, without the need for scale-like structure as other IP interrupt wait (to avoid unnecessary context switches). For the long-running batch operations for a particular transaction, etc., DB2 agent is necessary for the CPU time, and DB2 will automatically determine whether the dynamic yield the CPU time.
DB2 pureScale built-in load balancing for the cluster members is another important mechanism for scalability features of DB2. Do not have cluster-aware applications can use load balancing of the mechanism. DB2 for z / OS data sharing customers are now used by the client driver for DB2 pureScale feature provides load balancing cluster.
Scale-out architectures of the role is not limited to increased capacity to retain resources. Systems designed using this architecture in the face of component failure can continue to process transactions, which can deliver higher availability.
And distributed platforms compared to other products, DB2 pureScale the availability to upgrade to a new height. DB2 pureScale do not allow access to all the data recovery page, and pages ready for insight into what needs to be restored, without the need to perform any I / O operations. This is a unique feature of CF through centralized to achieve another important innovation.
Whenever members of the paging buffer pool read it when, CF will be aware of this incident and continue to track it. As long as members want to update the distribution lines, CF-related events also can be known. When an application performs a transaction, members of the dirty pages will be written directly to the CF memory. This process allows the cluster want to read the changed pages of any other member to get updates directly from the CF. More importantly, from the perspective of recovery, if any member fails, CF members will be present in the failure of some of the pages are being updated, along with some of the pages have been updated and submitted to complete, but has not yet written to disk.
Any relational database management system (RDBMS) of the recovery process first of all need to perform any committed transactions to ensure that these matters on the disk paging is the latest (this process is called redo recovery). In addition, any database server need to withdraw any pending transactions that failed disk data before the implementation of a change but not yet submitted (this process is called the revocation of recovery).
In the shared disk cluster, ensure that no other nodes in the cluster have not yet read or update any recovery disk page (you can restore the pages of these lines after the implementation of new services). This is the CF flash point: As the CF to know which page is in the process of updating the failed node, and CF has been submitted to the node the dirty pages stored in its centralized buffer pool, so DB2 pureScale in determining the Which pages need to restore the other members do not have to continue blocking the transaction. Other architectures will need to know which nodes take up more processing time to the distribution of information according to the lock to determine which nodes must be restored (see below).
From a high level, can easily explain the DB2 pureScale environment in this recovery process. Each member has a free process, but they are ready to deal with failure events. When a member fails, one of the recovery process will be activated; these processes already exist, so the operating system does not have to waste valuable system time to create a process, such as allocating memory for it. This recovery process will immediately CF dirty pages in the prefetch buffer to its own local pool. Most of the recovery process do not require I / O operations, because of the need to restore the page has been concentrated in the buffer pool of CF. In addition, the page prefetch mechanism will use the lightweight RDMA and recovery in the CF members to achieve rapid and effective transmission. During this time, all the other members of the all the other applications will continue to process the request. If they need to recover from any pages without access to any data, so they can
Continue to implement their own affairs. Therefore, they can continue to read the page from disk, as CF already know what pages on the disk is clean, and which pages need to restore. Then, members of the failure recovery process reads the log file, so replay is necessary to redo the transaction or cancel the updates made to the members fault.
For a typical transaction workload, from the member fails to complete the failed node does not update the page for other services usually use intervals of 20 seconds. Note that this also includes fault detection time, and some suppliers that the recovery time when they ruled this time. Database has kept all the other pages (even after the failure of the members) are fully available.
In addition, the system PowerHA pureScale cluster accelerator such as the components are redundant. The dual function of supporting CF DB2 pureScale, so lock and share cached information can be stored in two separate locations, in response to failure of the main CF.
Through the use of modern hardware architecture, DB2 pureScale could previously only in DB2 for z / OS available on a centralized lock and introduced into the distributed caching platform. The use of hardware and network to improve the level of concurrency and significantly reduce the overhead, thus providing a higher level of scalability. In addition, focus lock, and page caching allows DB2 pureScale continued perception of failure encountered in the members page which needs to be restored. Therefore, in the face of failure, all data is still not restored for use by other applications, while the failed node are updated page will be more quickly restored.
Need for high availability and achieve cost benefits through the horizontal development of applications, DB2 pureScale provides a solution to meet these requirements and has passed the market test.