Informix Dynamic Server 11: Advanced Functionality for Modern Business

4 Informix Dynamic Server 11: Advanced Functionality for Modern Business

1.1.1 DSA components: Processor

IDS provides the unique ability to scale the data system by employing a

dynamically configurable pool of data server processes called

Virtual Processors

(VPs). Database operations, such as a sorted data query, are segmented into

task-oriented subtasks (data read, join, group, and sort, as examples) for rapid

processing by Virtual Processors that specialize in that type of subtask. Virtual

Processors mimic the functionality of the hardware CPUs in that those VPs

schedule and manage user requests using multiple, concurrent threads. This is

illustrated in Figure 1-1.

Figure 1-1 IDS pool of VPs for parallel task execution

A thread represents a discrete task within a data server process and many

threads can execute simultaneously, and in parallel, across the pool of Virtual

Processors. Unlike a CPU process-based (or single-threaded) engine, which

leaves tasks on the system CPU for its given unit of time (even if no work can be

done, and thus wasting processing time), Virtual Processors are multi-threaded.

Consequently, when a thread is either waiting for a resource or has completed its

task, a thread switch occurs and the Virtual Processor immediately works on

another thread. As a result, precious CPU time is not only saved, but it is used to

satisfy as many user requests as possible in the given amount of time. This

CPU-1 CPU-2 CPU-3 CPU-4

CPU AIO

ENCRYPT

UD-VP

Virtual processors

Shared Memory

Buffers

Session

Info

SQL

Cache

Message

Chapter 1. IBM Informix Dynamic Server 11 essentials 5

enables an efficient use of hardware resources and is referred to as

fan-in

parallelism

. It is illustrated in Figure 1-2.

Figure 1-2 Fan-in parallelism

Not only can one Virtual Processor respond to multiple user requests in any

given unit of time, but one user request can also be distributed across multiple

Virtual Processors. For example, with a processing-intensive request, such as a

multi-table join, the data server divides the task into multiple subtasks and then

spreads these subtasks across all available Virtual Processors. Fan-out

parallelism uses many VPs to process a single SQL operation. With the ability to

distribute tasks, the request is completed quicker. This is referred to as

fan-out

parallelism

and is illustrated in Figure 1-3.

Figure 1-3 Fan-out parallelism

Virtual Processor

Processor

Physical

User 1 User 2 User 3 User 4 User 5

Processor 1

Physical

Processor 2

Physical

Processor 3

Physical

Virtual Processor 1 Virtual Processor 2 Virtual Processor 3

6 Informix Dynamic Server 11: Advanced Functionality for Modern Business

Together with fan-in parallelism, the net effect is more work being accomplished

quicker than with single-threaded architectures; in other words, the data server is

faster.

Dynamic load balancing occurs within IDS, because threads are not statically

assigned to Virtual Processors. Outstanding requests are serviced by the first

available Virtual Processor, balancing the workload across all available

resources. For efficient execution and versatile tuning, Virtual Processors can be

grouped into classes, each of which is optimized for a particular function, such as

CPU operations, disk I/O, communications, and administrative tasks, as

illustrated in Figure 1-4.

Figure 1-4 VPs are grouped into classes and optimized by function

An administrator can configure the instance with the appropriate number of

Virtual Processors in each class to handle the workload. Adjustments can be

made while the instance is online without interrupting database operations in

order to handle occasional periods of heavy activity or different load mixes. With

IDS 11, the data server in certain cases automatically adjusts the mix of VPs to

meet the current workload.

In UNIX and Linux systems, the use of multi-threaded Virtual Processors

significantly reduces the number of UNIX and Linux processes and,

consequently, less context switching is required. In Microsoft Windows systems,

Dynamically tunable

Parallelized, multi-threaded,

data server operating system

Faster context switching

Better scheduling

Better locking

Dynamic memory

CPU VP

Communication VPs

TCP/IP, Shared Memory, etc.

AIO VP

Administration VP

User-defined VP

Spatial VP

IDS Data Server