Linux cluster

The literature/X-ray data was sparse (-1000), so the information content was amplified by generating vast numbers of alternative conformations (-10 ) with a 64-node Linux cluster. Typically, we were able to generate 1 distinct conformation per compound in an average of <1 per minute. At critical points during the drug discovery process, such as when new structural data was obtained, we retrained our free energy model to improve its performance. 

Van Vliet et al. (2005) tracked 1x10 computational nodes to obtain a stochastic solution of their FDF equations. In order to deal with the high computational costs, the code was run in parallel on a Linux cluster of 11 dual AMD Athlon (TM) MP 1800 +processors. In this way, about one turbulent macro time scale (or 8,000 computational steps) per 2 days was computed. 

CellProfiler is designed to analyze large amounts of images automatically (17, 18). Since the analysis of thousands of images is very computationally intensive, a version is available to run on a Linux cluster allowing the parallelization of the analysis. The software is written in the Python programming language and the source code can be downloaded and extended by the community. 

We analyze our images with CellProfiler installed on a Linux cluster to speed up this calculation-intensive process. The cluster is a 17-node cluster (1 master node, 16 compute nodes). Each node has 8 cores with 2.26 GHz and 48 GB of RAM. The master node shares 24 TB of disk space out to the compute nodes over NFS. 

For data evaluation also self-programmed software (mostly in BTCL, Fortran, C) was applied. Data production runs were performed on a 74 processor Opteron Linux Cluster, a SGI Origin 2100 and on SGI Onyx workstation. 

This research was partially by NSF CAREER Award DBE0237901 INGEN (Indiana Genomics Initiatives), and AVIDD (Analysis and Visualization of Instrument-Driven Data) Linux cluster. 

Figure 17.4 shows the daily log of one node of the system from the 100-node LINUX cluster. This graph shows the number of transactions per second on the... 

FIGURE 17.4 LINUX cluster performance for 24 hours. The log of transactions per minute for single element of 100-node cluster. 

The LINUX cluster had been managed under continually increasing load for a period of months. Systems administrators had responded by incremental addition of PCs, but the performance had shown continued decline. The system was loaded near capacity, which is often far below theoretical, 70% of maximum being a common benchmark. Another feature of the system, which is especially relevant to clusters, was the impact of component failure. For computers, this is likely the disk drive. Drives have a mean time between failure of one to two years and follow a bathtub curve of burn-in failure, followed by reliable performance and then high failure near the end of the lifecycle. Cluster design must take into account the need for graceful degradation. 

Wayne, J. S and A. Mink. 1999. LINUX clusters at NIST. LINUX Journal 62 105-109. 

The extended simple point charge (SPC/E) model [59] is used. This model is known to give reasonably accurate values of static dielectric permittivity of liquid water at ambient conditions [60]. The MD simulations were performed for both H2O and D2O with the system size of 1024 particles at 220 K, 240 K, 267 K, 273 K, 300 K, and 355 K. The parallel molecular dynamics code for arbitrary molecular mixtures (DynaMix) is implemented by Lyubartsev and Laaksonen [61]. The simulations have been carried out on a Linux cluster built on the Tyan/Opteron 64 platform, which enables calculations of relatively long trajectories for a system of 1024 water molecules. The simulation run lengths depend on temperature and are in the range between 1 ns and 4 ns for the warmest and coldest simulation, respectively. As the initial condition was a cubic lattice, the equilibration time was chosen to be temperature dependent in the range from 200 ps at 355 Ktol ns at 200K. 

Average bandwidth obtained with adaptive and static routing, shown as a fraction of the peak bandwidth measured for each case. InfiniBand with 4x single data rate was used for static routing, and 10 Gigabit Ethernet was used for adaptive routing. Data were obtained on a Linux cluster and represent average measured bandwidths for the Sandia Cbench Rotate benchmark. ... 

Latency a, inverse bandwidth p, and bandwidth for Gigabit Ethernet (GigE) and InfiniBand (using IPoIB) interconnects on a Linux cluster. Data were determined using the program shown in Figure 5.2, and the reported bandwidths are unidirectional... 

To determine values for the machine parameters a, p, and y, a series of test runs were performed using a Linux cluster. The value for y was estimated to be 4.3 ns by timing single-process matrix-vector multiplications for various matrix sizes. To model the communication time, the values of a and the sum ip + y are required these values were found to be a = 43 jxs and 2p + Y = 72 ns/word (using 8 byte words) by timing the all-reduce operation as a function of the number of processes for a number of problem sizes and fitting the data to a function of the form of Eq. 5.20. Using these values for the machine parameters, the performance model was used... 

Speedups for MP2 algorithms PI and P2 (dynamic version) measured relative to timings for one process (Pli and P2i) and sixteen processes (Plie and P2i6). Computations tvere performed on a Linux cluster for the uracil dimer molecule using the cc-pVDZ basis set (cf. Figure 9.8). Inflated speedup curves are obtained by measuring speedups relative to a number of processes greater than one. 

Superlinear speedups for Fock matrix formation in the iterative part of the Harlree-Fock procedure as a consequence of storing a larger fraction of the integrals on each process as the number of processes increases. Speedups for the entire Hartree-Fock procedure are shown as well. Computations were performed on a Linux cluster for the uracil dimer using the aug-cc-pVTZ basis set (cf. Figure 8.3). A static task distribution of atom quartets was employed (see section 8.3 for details of the algorithm). 

The average time n) and the ratio of the standard deviation to the average time cr/fi) for computation of shell quartets of two-electron integrals. Results were obtained for the ethane molecule on a single processor of a Linux cluster. The number of basis functions and shells in the basis set are denoted n and Wsheii/ respectively. The subscript q refers to individual shell quartets of integrals, (MN RS), whereas the subscript p refers to sets of integrals (MN RS) for one MN pair and all included RS pairs... 

Predicted and measured parallel efficiencies on a Linux cluster for two-electron integral computation for alkanes employing the cc-pVTZ basis set and using static distribution of shell quartets (a) and shell pairs (b). ... 

A Linux cluster consisting of nodes with two single-core 3.06 GHz Intel Xeon processors (each with 512 KiB of L2 cache) connected via a 4x Single Data Rate InfiniBand network with a full fat tree topology. 

Speedups for parallel Fock matrix formation using fully distributed density and Fock matrices. Speedups were obtained using one compute thread per node on a Linux cluster for the uracil dimer with the aug-cc-pVTZ basis set. Speedups were computed relative to single-process... 

Log-log plot of predicted total execution times (Pltotal and P2totai) and communication times (Plcomm and P2comm) on a Linux cluster for the uracil dimer using the cc-pVDZ basis set. Ideal scaling corresponds to a straight line of negative unit slope. 

The measured total execution times and communication times for PI and P2 on the employed Linux cluster" are illustrated in Figure 9.7. For larger... 

---