11-02-2010, 08:21 PM
CLUSTERCOMPUTING.doc (Size: 178 KB / Downloads: 353)
ABSTRACT
Not all difficult problems require access to a single shared memory resource. Some problems can easily be broken into many smaller independent parts. Computer scientists often refer to this class of problems as "embarrassingly parallel" or as capacity problems. Many of the computers that we typically employ on a day-to-day basis for word processing or for game playing are very well equipped to solve the smaller components of capacity problems. In practice, clusters are usually composed of many commodity computers, linked together by a high-speed dedicated network
What distinguishes this configuration from the heavy hitting, top dollar supercomputers is that each node within a cluster is an independent system, with its own operating system, private memory, and, in some cases, its own file system. Because the processors on one node cannot directly access the memory on the other nodes, programs or software run on clusters usually employ a procedure called "message passing" to get data and execution code from one node to another. Compared to the shared memory systems of supercomputers, passing messages is very slow.
However, with cluster computing, the subparts of the larger problem usually run on a single processor for a long period of time without reference to the other sub parts, which means that the slow communication among nodes is not a major problem. Experts in the field often refer to these types of problems as CPU-bound. Cluster computing has become a major part of many research programs because the price to performance ratio of commodity clusters is very good. Also, because the nodes in a cluster are clones, there is no single point of
failure, which enhances the reliability to the cluster. Of course, these benefits can only be realized if the problems you are attempting to solve can be easily parallelized. Increasingly, computer clusters are being combined with large shared memory systems, such as the ones found in supercomputing architectures. By doing so, scientists who work on problems that have both capability and capacity elements can take advantage of the inherent strengths of both designs.
1. INTRODUCTION
Parallel computing has seen many changes since the days of the highly expensive and proprietary supercomputers. Changes and improvements in performance have also been seen in the area of mainframe computing for many environments. But these compute environments may not be the most cost effective and flexible solution for a problem.
Over the past decade, cluster technologies have been developed that allow multiple low cost computers to work in a coordinated fashion to process applications. The economics, performance and flexibility of compute clusters makes cluster computing an attractive alternative to centralized computing models and the attendant to cost, inflexibility, and scalability issues inherent to these models.
Many enterprises are now looking at clusters of high-performance, low cost computers to provide increased application performance, high availability, and ease of scaling within the data center. Interest in and deployment of computer clusters has largely been driven by the increase in the performance of off-the-shelf commodity computers, high-speed, low-latency network switches and the maturity of the software components.
Application performance continues to be of significant concern for various entities including governments, military, education, scientific and now enterprise organizations. This document provides a review of cluster computing, the various types of clusters and their associated applications. This document is a high-level informational document; it does not provide details about various cluster implementations and applications.
2. CLUSTER COMPUTING
Cluster computing is best characterized as the integration of a number of off-the-shelf commodity computers and resources integrated through hardware, networks, and software to behave as a single computer. Initially, the terms cluster computing and high performance computing were viewed as one and the same. However, the technologies available today have redefined the term cluster computing to extend beyond parallel computing to incorporate load-balancing clusters (for example, web clusters) and high availability clusters. Clusters may also be deployed to address load balancing, parallel processing, systems management, and scalability.
Today, clusters are made up of commodity computers usually restricted to a single switch or group of interconnected switches operating at Layer 2 and within a single virtual local-area network (VLAN).Each compute node (computer) may have different characteristics such as single processor or symmetric multiprocessor design, and access to various types of storage devices. The underlying network is a dedicated network made up of high-speed, low-latency switches that may be of a single switch or a hierarchy of multiple switches. A growing range of possibilities exists for a cluster interconnection technology. Different variables will determine the network hardware for the cluster. Price-per-port, bandwidth, latency, and throughput are key variables. The choice of network technology depends on a number of factors, including price, performance, and compatibility with other cluster hardware and system software as well as communication characteristics of the applications that will use the cluster.
Clusters are not commodities in themselves, although they may be based on commodity hardware. A number of decisions need to be made (for example, what type of hardware the nodes run on, which interconnect to use, and which type of switching architecture to build on) before assembling a cluster range. Each decision will affect the others, and some will probably be dictated by the
intended use of the cluster. Selecting the right cluster elements involves an understanding of the application and the necessary resources that include, but are not limited to, storage, throughput, latency, and number of nodes.
When considering a cluster implementation, there are some basic questions that can help determine the cluster attributes such that technology options can be evaluated:
1. Will the application be primarily processing a single dataset
2. Will the application be passing data around or will it generate real-time information
3. Is the application 32- or 64-bit
The answers to these questions will influence the type of CPU, memory architecture, storage, cluster interconnect, and cluster network design. Cluster applications are often CPU-bound so that interconnect and storage bandwidth are not limiting factors, although this is not always the case.
3. CLUSTER BENEFITS
The main benefits of clusters are scalability, availability, and performance. For scalability, a cluster uses the combined processing power of compute nodes to run cluster-enabled applications such as a parallel database server at a higher performance than a single machine can provide. Scaling the cluster's processing power is achieved by simply adding additional nodes to the cluster.
Availability within the cluster is assured as nodes within the cluster provide backup to each other in the event of a failure. In high-availability clusters, if a node is taken out of service or fails, the load is transferred to another node (or nodes) within the cluster. To the user, this operation is transparent as the applications and data running are also available on the failover nodes.
An additional benefit comes with the existence of a single system image and the ease of manageability of the cluster. From the users perspective the users sees an application resource as the provider of services and applications. The user does not know or care if this resource is a single server, a cluster, or even which node within the cluster is providing services.
These benefits map to needs of today's enterprise business, education, military and scientific community infrastructures. In summary, clusters provide:
¢ Scalable capacity for compute, data, and transaction intensive applications, including support of mixed workloads
Horizontal and vertical scalability without downtime
¢ Ability to handle unexpected peaks in workload
¢ Central system management of a single systems image
¢ 24 x 7 availability
4. TYPES OF CLUSTERS
There are several types of clusters, each with specific design goals and functionality. These clusters range from distributed or parallel clusters for computation intensive or data intensive applications that are used for protein, seismic, or nuclear modeling to simple load-balanced clusters.
4.1 High Availability or Failover Clusters
These clusters are designed to provide uninterrupted availability of data or services (typically web services) to the end-user community. The purpose of these clusters is to ensure that a single instance of an application is only ever running on one cluster member at a time but if and when that cluster member is no longer available, the application will failover to another cluster member. With a high-availability cluster, nodes can be taken out-of-service for maintenance or repairs. Additionally, if a node fails, the service can be restored without affecting the availability of the services provided by the cluster (see Figure 1). While the application will still be available, there will be a performance drop due to the missing node.
High-availability clusters implementations are best for mission-critical applications or databases, mail, file and print, web, or application servers.
Normal operation
Nodel Node2
Heart Beat
ApplnA « ApplnB
Shared Storage
After Fail Over
Nodel Heart Beat Node2
X 4 Appln A
« ApplnB
Shared Storaee
Figure 1
Unlike distributed or parallel processing clusters, high-availability clusters seamlessly and transparently integrate existing standalone, non-cluster aware applications together into a single virtual machine necessary to allow the network to effortlessly grow to meet increased business demands.
4.2 Clusters-Aware and Cluster-Unaware Applications
Cluster-aware applications are designed specifically for use in clustered environment. They know about the existence of other nodes and are able to communicate with them. Clustered database is one example of such application. Instances of clustered database run in different nodes and have to notify other instances if they need to lock or modify some data.
Cluster-unaware applications do not know if they are running in a cluster or on a single node. The existence of a cluster is completely transparent for such applications, and some additional software is usually needed to set up a cluster. A web server is a typical cluster-unaware application. All servers in the cluster have the same content, and the client does not care from which server the server provides the requested content.
4.3 Load Balancing Cluster
This type of cluster distributes incoming requests for resources or content among multiple nodes running the same programs or having the same content (see Figure 2). Every node in the cluster is able to handle requests for the same content or application. If a node fails, requests are redistributed between the remaining available nodes. This type of distribution is typically seen in a web-hosting environment.
Both the high availability and load-balancing cluster technologies can be combined to increase the reliability, availability, and scalability of application and data resources that are widely deployed for web, mail, news, or FTP services.
4.4 Parallel/Distributed Processing Clusters
Traditionally, parallel processing was performed by multiple processors in a specially designed parallel computer. These are systems in which multiple processors share a single memory and bus interface within a single computer. With the advent of high speed, low-latency switching technology, computers can be interconnected to form a parallel-processing cluster. These types of cluster increase availability, performance, and scalability for applications, particularly computationally or data intensive tasks.
A parallel cluster is a system that uses a number of nodes to simultaneously solve a specific computational or data-mining task. Unlike the load balancing or high-availability cluster that distributes requests/tasks to nodes where a node processes the entire request, a parallel environment will divide the request into multiple sub-tasks that are distributed to multiple nodes within the cluster for processing. Parallel clusters are typically used for CPU-intensive analytical applications, such as mathematical computation, scientific analysis (weather forecasting, seismic analysis, etc.), and financial data analysis.
One of the more common cluster operating systems is the Beowulf class of clusters. A Beowulf cluster can be defined as a number of systems whose collective processing capabilities are simultaneously applied to a specific technical, scientific, or business application. Each individual computer is referred to as a "node" and each node communicates with other nodes within a cluster across standard Ethernet technologies (10/100 Mbps, GibeE, or 10GbE). Other high-speed interconnects such as Myrinet, Infiniband, or Quadrics may also be used.
5. CLUSTER COMPONENTS
The basic building blocks of clusters are broken down into multiple categories: the cluster nodes, cluster operating system, network switching hardware and the node/switch interconnect. Significant advances have been accomplished over the past five years to improve the performance of both the compute nodes as well as the underlying switching infrastructure.
5.1 Cluster Nodes
Node technology has migrated from the conventional tower cases to single rack-unit multiprocessor systems and blade servers that provide a much higher processor density within a decreased area. Processor speeds and server architectures have increased in performance, as well as solutions that provide options for either 32-bit or 64-bit processors systems. Additionally, memory performance as well as hard-disk access speeds and storage capacities have also increased. It is interesting to note that even though performance is growing exponentially in some cases, the cost of these technologies has dropped considerably.
As shown in Figure 3 below, node participation in the cluster falls into one of two responsibilities: master (or head) node and compute (or slave) nodes. The master node is the unique server in cluster systems. It is responsible for running the file system and also serves as the key system for clustering middleware to route processes, duties, and monitor the health and status of each slave node. A compute (or slave) node within a cluster provides the cluster a computing and data storage capability. These nodes are derived from fully operational, standalone computers that are typically marketed as desktop or server systems that, as such, are off-the-shelf commodity systems.
Computing Nodes
5.2 Cluster Network
Commodity cluster solutions are viable today due to a number of factors such as the high performance commodity servers and the availability of high speed, low-latency network switch technologies that provide the inter-nodal communications. Commodity clusters typically incorporate one or more dedicated switches to support communication between the cluster nodes. The speed and type of node interconnects vary based on the requirements of the application and organization.
With today's low costs per-port for Gigabit Ethernet switches, adoption of 10-Gigabit Ethernet and the standardization of 10/100/1000 network interfaces on the node hardware, Ethernet continues to be a leading interconnect technology for many clusters. In addition to Ethernet, alternative network or interconnect technologies include Myrinet, Quadrics, and Infiniband that support bandwidths above 1Gbps and end-to-end message latencies below 10 microseconds (uSec).
5.3 Network Characterization
There are two primary characteristics establishing the operational properties of a network: bandwidth and delay. Bandwidth is measured in millions of bits per second (Mbps) and/or billions of bits per-second (Gbps). Peak bandwidth is the maximum amount of data that can be transferred in a single unit of time through a single connection. Bi-section bandwidth is the total peak bandwidth that can be passed across a single switch.
Latency is measured in microseconds (uSec) or milliseconds (mSec) and is the time it takes to move a single packet of information in one port and out of another. For parallel clusters, latency is measured as the time it takes for a message to be passed from one processor to another that includes the latency of the interconnecting switch or switches. The actual latencies observed will vary widely even on a single switch depending on characteristics such as packet size, switch architecture (centralized versus
distributed), queuing, buffer depths and allocations, and protocol processing at the nodes.
5.4 Ethernet, Fast Ethernet, Gigabit Ethernet and 10-Gigabit Ethernet
Ethernet is the most widely used interconnect technology for local area networking (LAN). Ethernet as a technology supports speeds varying from 10Mbps to 10 Gbps and it is successfully deployed and operational within many high-performance cluster computing environments.
6. CLUSTER APPLICATION
Parallel applications exhibit a wide range of communication behaviors and impose various requirements on the underlying network. These may be unique to a specific application, or an application category depending on the requirements of the computational processes.
Some problems require the high bandwidth and low-latency capabilities of today's low-latency, high throughput switches using 10GbE, Infiniband or Myrinet. Other application classes perform effectively on commodity clusters and will not push the bounds of the bandwidth and resources of these same switches. Many applications and the messaging algorithms used fall in between these two ends of the spectrum.
Currently, there are four primary categories of applications that use parallel clusters: compute intensive, data or input/output (I/O) intensive, and transaction intensive. Each of these has its own set of characteristics and associated network requirements. Each has a different impact on the network as well as how each is impacted by the architectural characteristics of the underlying network. The following subsections describe each application types.
6.1 Compute Intensive Application
Compute intensive is a term that applies to any computer application that demands a lot of computation cycles (for example, scientific applications such as meteorological prediction). These types of applications are very sensitive to end-to-end message latency. This latency sensitivity is caused by either the processors having to wait for instruction messages, or if transmitting results data between nodes takes longer. In general, the more time spent idle waiting for an instruction or for results data, the longer it takes to complete the application.
Some compute-intensive applications may also be graphic intensive. Graphic intensive is a term that applies to any application that demands a lot of computational cycles where the end result is the delivery of significant information for the development of graphical output such as ray-tracing applications. These types of applications are also sensitive to end-to-end message latency. The longer the processors have to wait for instruction messages or the longer it takes to send resulting data, the longer it takes to present the graphical representation of the resulting data.
6.2 Data or I/O Intensive Applications
Data intensive is a term that applies to any application that has high demands of attached storage facilities. Performance of many of these applications is impacted by the quality of the I/O mechanisms supported by current cluster architectures, the bandwidth available for network attached storage, and, in some cases, the performance of the underlying network components at both Layer 2 and 3. Data-intensive applications can be found in the area of data mining, image processing, and genome and protein science applications. The movement to parallel I/O systems continues to occur to improve the I/O performance for many of these applications.
6.3 Transaction Intensive Applications
Transaction intensive is a term that applies to any application that has a high-level of interactive transactions between an application resource and the cluster resources. Many financial, banking, human resource, and web-based applications fall into this category.
7. PERFPRMANCE IMPACT AND CAREABOUTS
There are three main careabouts for cluster applications: message latency, CPU utilization, and throughput. Each of these plays an important part in improving or impeding application performance. This section describes each of these issues and their associated impact on application performance.
8. MESSAGE LATENCY
Message latency is defined as the time it takes to send a zero-length message from one processor to another (measured in microseconds). The lower the latency for some application types, the better. Message latency is made up of aggregate latency incurred at each element within the cluster network, including within the cluster nodes themselves .Although network latency is often focused on, the protocol processing latency of message passing interface (MPI) and TCP processes within the host itself are typically larger.
Throughput of today's cluster nodes are impacted by protocol processing, both for TCP/IP processing and the MPI. To maintain cluster stability, node synchronization, and data sharing, the cluster uses message passing technologies such as Parallel Virtual Machine (PVM) or MPI.
TCP/IP stack processing is a CPU-intensive task that limits performance within high speed networks. As CPU performance has increased and new techniques such as TCP offload engines (TOE) have been introduced, PCs are now able to drive the bandwidth levels higher to a point where we see traffic levels reaching near theoretical maximum for TCP/IP on Gigabit Ethernet and near bus speeds for PCI-X based systems when using 10
Gigabit Ethernet. These high-bandwidth capabilities will continue to grow as processor speeds increase and more vendors build network adapters to the PCI-Express specification.
To address host stack latency, reductions in protocol processing have been addressed somewhat through the implementation of TOE and further developments of combined TOE and Remote Direct Memory Access (RDMA) technologies are occurring that will significantly reduce the protocol processing in the host.
9. CPU UTILIZATION
One important consideration for many enterprises is to use compute resources as efficiently as possible. As increased number of enterprises move towards realtime and business-intelligence analysis, using compute resources efficiently is an important metric. However, in many cases compute resource is underutilized. The more CPU cycles committed to application processing the less time it takes to run the application. Unfortunately, although this is a design goal, this is not obtainable as both the application and protocols compete for CPU cycles.
As the cluster node processes the application, the CPU is dedicated to the application and protocol processing does not occur. For this to change, the protocol process must interrupt a uniprocessor machine or request a spin lock for a multiprocessor machine. As the request is granted, CPU cycles are then applied to the protocol process. As more cycles are applied to protocol processing, application processing is suspended. In many environments, the value of the cluster is based on the run-time of the application. The shorter the time to run, the more floating-point operations and/or millions of instructions per-second occur, and, therefore, the lower the cost of running a specific application or job.
Figure 4 CPU Utilization
Application
Application
Application processing
Spin lockp
Application in wait state
The example shows that when there is virtually no network or protocol processing going on, CPU 0 and 1 of each node are 100% devoted to application processing.
It also shows that the network traffic levels have significantly increased. As this happens, the CPU spends cycles processing the MPI and TCP protocol stacks, including moving data to and from the wire. This results in a reduced or suspended application processing. With the increase in protocol processing, note that the utilization percentages of CPU 0 and 1 are dramatically reduced, in some cases to 0.
10. THROUGHPUT
Data throughput begins with a calculation of a theoretical maximum throughput and concludes with effective throughput. The effective throughput available between nodes will always be less than the theoretical maximum. Throughput for cluster nodes is based on many factors, including the following:
¢ Total number of nodes running
¢ Switch architectures
¢ Forwarding methodologies
¢ Queuing methodologies
¢ Buffering depth and allocations
¢ Noise and errors on the cable plant
As previously noted, parallel applications exhibit a wide range of communication behaviors and impose various requirements on the underlying network. These behaviors may be unique to individual applications and the requirements for inter-processor/inter-nodal communication. The methods used by the application programmer, as far as the passing of messages using MPI, vary based on the application requirements.
There are both simple and complex collective routines. As more scatter-gather, all gather, and all-to-all routines are used, multiple head-of-line blocking instances may occur within the switch, even within non-blocking switch architectures. Additionally, the buffer architectures of the underlying network, specifically the depth and allocation of ingress and egress buffers, become key to throughput levels.
If buffers fill, congestion management routines may be invoked. In the switch, this means that pause frames will be sent resulting in the sending node discontinuing sending traffic until the congestion subsides. In the case of TCP, the congestion avoidance algorithms come into effect.
11. SLOW START
In the original implementation of TCP, as soon as a connection was established between two devices, they could each send segments as fast as they liked as long as there was room in the other devices receive window. In a busy network, the sudden appearance of a large amount of new traffic could exacerbate any existing congestion.
To alleviate this problem, modern TCP devices are restrained in the rate at which they initially send segments. Each sender is at first restricted to sending only an amount of data equal to one "full-sized" segment that is equal to the MSS value for the connection.
Each time an acknowledgment is received, the amount of data the device can send is increased by the size of another full-sized segment. Thus, the device "starts slow" in terms of how much data it can send, with the amount it sends increasing until either the full window size is reached or congestion is detected on the link. In the latter case, the congestion avoidance feature, described below, is used.
12. CONGESTION AVOIDANCE
When potential congestion is detected on a TCP link, a device responds by throttling back the rate at which it sends segments. A special algorithm is used that allows the device to drop the rate at which segments are sent quickly when congestion occurs. The device then uses the Slow Start algorithm, described above, to gradually increase the transmission rate back up again to try to maximize throughput without congestion occurring again.
In the event of packet drops, TCP retransmission algorithms will engage. Retransmission timeouts can reach delays of up to 200 milliseconds, thereby significantly impacting throughput.
13. SUMMARY
High-performance cluster computing is enabling a new class of computationally intensive applications that are solving problems that were previously cost prohibitive for many enterprises. The use of commodity computers collaborating to resolve highly complex, computationally intensive tasks has broad application across several industry verticals such as chemistry or biology, quantum physics, petroleum exploration, crash test simulation, CG rendering, and financial risk analysis. However, cluster computing pushes the limits of server architectures, computing, and network performance.
Due to the economics of cluster computing and the flexibility and high performance offered, cluster computing has made its way into the mainstream enterprise data centers using clusters of various sizes.
As clusters become more popular and more pervasive, careful consideration of the application requirements and what that translates to in terms of network characteristics becomes critical to the design and delivery of an optimal and reliable performing solution.
Knowledge of how the application uses the cluster nodes and how the characteristics of the application impact and are impacted by the underlying network is critically important. As critical as the selection of the cluster nodes and operating system, so too are the selection of the node interconnects and underlying cluster network switching technologies.
A scalable and modular networking solution is critical, not only to provide incremental connectivity but also to provide incremental bandwidth options as the cluster grows. The ability to use advanced technologies within the same networking platform, such as 10 Gigabit Ethernet, provides new connectivity options, increases bandwidth, whilst providing investment protection.
The technologies associated with cluster computing, including host protocol stack-processing and interconnect technologies, are rapidly evolving to meet the demands of current, new, and emerging applications. Much progress has been made in the development of low-latency switches, protocols, and standards that efficiently and effectively use network hardware components.
1. Introduction 1
2. Cluster Computing
3. Cluster Benefits 4
4. Types of Clusters 5
4.1 High Availability or Failover Clusters 5
4.2 Cluster-Aware and Cluster-Unaware Applications
4.3 Load Balancing Cluster
4.4 Parallel/Distributed Processing Clusters 8
5. Cluster Components 9
5.1 Cluster Nodes 9
5.2 Cluster Network 10
5.3 Network Characterization 11
5.4 Ethernet, Fast Ethernet 12
6. Cluster Applications 13
6.1 Compute Intensive Applications 13
6.2 Data or I/O Intensive Applications 13
6.3 Transaction Intensive Applications 14
7. Performance Impacts and Careabouts 15
8. Message Latency 15
9. CPU Utilization 17
10. Throughput 19
11. Slow Start 20
12. Congestion Avoidance 21
13. Summary 22