Scientific Computing Associates
has brought two of its high-performance computing products to Mac OS X.
The company’s Paradise middleware product has been brought to Mac OS X to enable distributing networking computing for users of the new operating system.
“We’re pleased to make Paradise available to Mac OS X users, extending productivity and increasing the return on investments,” said Beverly Thalberg, CEO and president of Scientific Computing Associates. “Idle CPU cycles can be recycled and aggregated to provide a ‘free’ virtual supercomputer for groups.”
Paradise is used to create high performance ensemble applications by enabling the coordination of separate, independent, computational processes through real-time sharing of data in its Virtual Shared Memory (VSM). Paradise applications can run on different types of operating systems and hardware, including Macs. It can be used to assemble applications from separate independent programs that may be written in different high-level languages such as C, C++, Java and Python.
For example, data from a mainframe database application might be placed into a VSM either as it’s collected or on a periodic basis. Then a separate visualization application on a Mac could retrieve the data from the VSM and display it graphically.
The two apps could run concurrently, in which case the visualization application would provide a live snapshot of the data. But they could also run at completely separate times, with the visualization application displaying summary information about data gathered from the previous day, for example.
Paradise features include the P2P Piranha system, which delivers performance and throughput gains by tapping into the huge reserve of computing power that’s available on most networks and multiprocessors. Piranha utilizes all available system capacity, Thalberg said. As tasks come and go, resources are quickly reallocated automatically and transparently to meet the changing demand, she added.
Piranha lets P2P distributed programs take advantage of idle cycles of machines without interfering with other uses of the systems, including interactive use, Thalberg said. At any given time, Piranha “declares” computers as either idle or busy, depending on user-definable criteria such as load average, keyboard or mouse activity.
When a machine becomes idle, it can join the Piranha computation and work on available tasks. When the system gets busy once more — say, if the user starts typing — the machine can immediately stop work on the Piranha computation and make its task available for another processor to finish.
What’s more, Scientific Computing Associates is bringing Linda to Apple’s new operating system. Linda, a “what” not a “she,” is an industry standard for parallel programming. Introduced in the mid-1980s, Linda was the first commercial product to implement Virtual Shared Memory (VSM) for supercomputers and workstations.
“Linda-enabled applications running on Mac clusters typically scale almost linearly,” Thalberg said. “Users can speed up their times to solution and also attack problems which have been intractable up to now.”
A Mac cluster is made up of a group of Macs interconnected by a fast network. The nodes of the cluster can be either desktop Macs or servers. Using Linda, you can split a program into multiple tasks running on different Macs, which synchronize and share data by means of the Linda VSM. When all the Macs in a cluster are running Mac OS X and Linda, you can take advantage of all the cluster’s resources in aggregate to “remarkably decrease” program execution time, Thalberg said.
The Linda shared memory model is content-addressable, not address based, making it much easier to build applications and fully utilize hardware capacity, Thalberg said. The Linda compiler provides support for both C and Fortran programming languages.
Linda applications include quantum chemistry, bioinformatics, risk analysis, simulations, data mining, portfolio optimization, seismic processing, design automation, ray tracing, engineering analysis and more.
Pricing for Paradise starts at US$3,000 for academic and $10,000 for commercial users. Pricing for Linda starts at $1,500 for academic and $5,000 for commercial users.
So, what’s the difference between Paradis and Linda? Jerry Lipchus, Scientific Computing Associates’ director of sales, told MacCentral that Linda is for “pedal to the medal” parallel computing as its multiple space is distributed among all the participants.
“Another big difference with Linda is that you can run only one application at a time, and Linda does not require a server,” he said. “It is licensed by the number of CPUs. We make a network of computers a supercomputer.”
Paradise is a client-server model in which the tuplespace is on a server and accessed by the clients. It’s more robust and general, supporting Windows as well as Unix and Linux, Lipchus said. It also supports more programming languages (such as Java and Python), has higher persistence and fault tolerance, and is optimized for distributed computing, he added.
With Paradise a server is required, and you can have multiple applications running at the same time. New applications and servers can be added on the fly. It’s licensed by servers and clients or connects.