Ericsson by: MCC International
Company: PlatformCustomer: EricssonSubmitted by: MCC InternationalDate: January 2000Large clusters of network servers can be utilised very efficiently with the help of a workload management system. With such a system, jobs are distributed so that all the servers are utilised optimally and with a minimum of wasted time.
Ericsson uses the workload management system called Load Sharing Facility from Platform Computing in parts of its operations both in Sweden and abroad.Having powerful processors and lots of memory in network servers does not always mean better performance for users. The more servers and programmes, and the more people and projects using them, the more difficult it becomes to utilise the resources in a network efficiently.Queues, bottlenecks and waiting time mean in reality that the resources are seldom as powerful as they appear to be on paper. This becomes especially apparent when resources are expanded prior to a particularly critical phase in a project, for example. The promise of an increase in performance all too often remains unfulfilled.
The problem often lies not in a lack of resources, but in a lack of co-ordination of these resources. This is because computers are often viewed as a static resource, while the demand for them is, in fact, dynamic. And the resources are not distributed on the basis of this continuously changing need.It is difficult to get a clear overview of the users’ needs and the availability of computing resources in a network. Unfortunately, it is all too common that no-one takes the trouble to analyse and solve this problem.
Instead, users select themselves which server they are going to run their jobs on.Naturally, they make their choices based on their own logic and perceptions of the appropriate level of computer power; choosing to run their calculation programme from their desktop system, and running simulations on the most powerful server in the network for speed. But often, the most powerful server is used for tasks that actually can be run in a computer with less capacity. This tends to create resource problems for the networked computer systems as a whole.Users make such choices independently of each other, since no-one can know what jobs their colleagues are thinking of running.
When several users want to run capacity-hungry jobs on the fastest machine, it easily becomes overburdened. The jobs in the machines are switched back and forth between the processors and disks as they try to share the available capacity, resulting in all the jobs running more slowly. Meanwhile, the queue of waiting jobs continues to grow. The fastest machine has suddenly become a bottleneck as a result of the lack of co-ordination of needs and resources. When each individual user tries to utilise the available resources optimally for his/her jobs, the system as a whole can easily become sub-optimised.
This phenomenon is beginning to become a serious problem, particularly for technology-intensive companies.Ericsson Utvecklings AB designs computerised circuits for Ericsson’s AXE exchanges. These circuits must be designed and verified with the help of very advanced programmes that simulate the circuits’ function. It is vital to utilise the available computer power efficiently, and to ensure that each short simulation which can take something between a few seconds and a few minutes is immediately followed by the next to reduce time wasted as far as possible. Sometimes simulations are much longer.
Then concepts like fault tolerance are important.Ericsson Utvecklings AB has solved these problems with the help of a workload management system. The system is called Load Sharing Facility (LSF), designed by the Canadian company Platform. Using LSF, Ericsson Utvecklings AB can get a clear overview of how the available resources are being utilised, and set up different queues for jobs to be run on different servers in the network. The queuing system is designed to provide queues for different jobs according to set priorities and the jobs’ demands on the computing resources.
Instead of letting the user choose which machine to run his/her job on, he/she chooses which queue to place the job in.LSF optimises the number of jobs being run in a server to ensure that all jobs are run as fast as possible. When the job is finished, LSF ensures that the server immediately starts new jobs.With the help of LSF, personnel at Ericsson Utvecklings AB can utilise all the servers in the network for simulations. LSF even utilises ordinary desktop systems for simulations when they are not being used for office work.
But if the user wants to have his/her desktop system back, LSF can immediately throw out the job being run on the desktop system, and start running it on another machine instead. Even though this results in a few jobs needing to be re-run, the collective utilisation of the resources in the cluster increases using LSF by more than 50 percent. LSF has been implemented at Ericsson Utvecklings AB in a computer environment made up exclusively of Sun Microsystems machines, thereby optimising the environment to become an exceptionally powerful design environment. Most of the clusters within Ericsson Utvecklings AB are based on Sun machines.Platform’s Swedish distributor, M;P HiTech Computing AB, participated actively in the qualification of the solution at Ericsson Utvecklings AB.
An important component of this solution was to ensure that LSF was integrated effectively with Ericsson Utvecklings AB’s business-critical applications.Barbara Remp, Manager EDA/R&D at Ericsson Utvecklings AB, says that “By using LSF, we have considerably improved the utilisation of our compute resources including both hardware and software. This means that we get much shorter turnaround times for computer jobs and thereby a considerable increase in the number of these jobs we can do in a fixed period of time.”Last autumn, Ericsson and Platform signed a framework agreement that means that all Ericsson’s development departments will be able to implement LSF for workload management of clusters. This means that LSF can now be used by Ericsson world-wide.
M;P HiTech Computing AB also assisted in drawing up the agreement with Ericsson. M;P has already delivered LSF to several Ericsson companies in Sweden, for example Ericsson Radio Systems AB, Ericsson Telecom AB and, of course, Ericsson Utvecklings AB.
