28 Sep Clusters and Clustering Methods
Please read through the following. The instructions are at the end.
Clusters and Clustering Methods
Clustering involves connecting multiple computers together in a network in a way that makes them behave like a single computer. This is another cost-effective approach for providing high performance. A cluster can often be built by simply using existing computers and workstations. To extend a cluster, you can simply add another computer to the network. Clustering is used to achieve parallel processing, load balancing, and fault tolerance.
Computer clusters can be classified into three main types, as follows (UKEssays, 2018):
• High-performance clusters
• High-availability clusters
• Load-balancing clusters
The most common and well-known benefits of clustering are as follows:
• Cost efficiency
• Incremental scalability
• High availability
• Fast processing speed
The following diagram shows two types of cluster configurations (Stallings, 2018):
There are five clustering methods, and the following table provides the description, benefits, and limitations of each clustering method (Stallings, 2018):
Non-Uniform Memory Access
Non-uniform memory access is a type of computer architecture that is often used in a symmetric multiprocessing (SMP) system. There are three main features associated with NUMA, as follows:
• It consists of a cluster of processors, and each processor has its own local cache memory that has direct access to the main memory through a commonly shared data communication bus.
• Each processor’s local cache memory can be accessed by all of the other processors within the same cluster through the same commonly shared data communication bus.
• The time and speed for each processor to access another processor’s local cache memory and the main memory are different to avoid conflict when using a commonly shared data communication bus.
Any application (e.g., a data mining or decision support system) can be partitioned into multiple tasks so that they can be concurrently processed by multiple processors and share a common database. Thus, a NUMA system is a good fit. Read this article on Non-Uniform Memory Access (NUMA) to learn more about why a NUMA system is a good fit.
One type of NUMA is a cache-coherent NUMA (ccNUMA), which is a NUMA system “in which cache coherence is maintained among the caches of various processors” (Stallings, 2018).
The following diagram shows a ccNUMA organization (Stallings, 2018):
The following table summarizes the pros and cons of a ccNUMA system (Stallings, 2018):
Pros Cons
A ccNUMA system can deliver effective performance at higher levels of parallelism than SMP without requiring major software changes. It does not transparently look like an SMP.
The bus traffic on an individual node is limited to a demand that the bus can handle. Software changes will be required to move an operating system and applications from an SMP to a ccNUMA system.
If many of the memory accesses are to remote nodes, performance begins to break down. There is some concern with its availability.
Assignment Details
Assume that you have been assigned the important task of helping a very large data center evaluate the following types of computer architectures for its data servers:
• Clustering
• NUMA
These data servers must also meet the following six requirements:
• Hardware: The data servers must use four 2.8 GHz Xeon multiprocessor (MP) processors, 8-way symmetric multiprocessing (SMP) that can be extended up to 16-way, 512 KB L2 cache, 2 MB L3 cache, and 2 GB (four pieces of 512 MB memory) main memory.
• Software: In addition to the operating system, the data server must be able to run on either the clustering software or the NUMA scheduler software.
• Networking: The data server must have the ability to communicate with remote nodes through the network interface.
• Application: These data servers must be able to run the main application: data mining based on relational databases.
• Availability: An availability of 99.99999% is required for every data server.
• Performance: The servers must have a high performance for data processing.
Your assignment is to create a short evaluation report after reviewing the six requirements above.
Your report must include the following:
• Title page
• Table of contents
• Introduction
• Pros and cons of clustering (including their justifications and applications)
• Pros and cons of NUMA (including their justifications and applications)
• Conclusion
• References
