National Partnership for Advanced Computational Infrastructure Advanced Architectures CSE 190 Reagan W. Moore San Diego Supercomputer Center

National Partnership for Advanced Computational Infrastructure Course Organization Professors / TA Sid Karin - Director, San Diego Supercomputer Center, Reagan Moore - Associate Director, SDSC Holly Dail - UCSD TA Seminars State of the art computer architectures Mid-term / SDSC tour Final exam

National Partnership for Advanced Computational Infrastructure Seminars 4/3 : Reagan Moore- Performance evaluation heuristics & modeling 4/10: Sid Karin - Historical perspective 4/17: Richard Kaufmann, Compaq - Teraflops systems 4/24: IBM or Sun 5/1 : Mark Seager, LLNL - ASCI 10 Tflops computer 5/8 : Midterm / SDSC Tour 5/15: John Feo, Tera - Multi-threaded architectures 5/22: Peter Beckman, LANL - Clusters 5/29: Holiday / no class 6/5 : Thomas Sterling, Caltech - Petaflops computers 6/12 : Final exam

National Partnership for Advanced Computational Infrastructure Distributed Archives Application Digital Library Data Mining Supercomputers for Simulation and Data Mining Information Discovery Collection Building

National Partnership for Advanced Computational Infrastructure Heuristics for Characterizing Supercomputers Generators of data - numerically intensive computing Usage models for the rate at which supercomputers move data between memory, disk, and archives Usage models for capacity of the data caches (memory size, local disk, and archival storage) Analyzers of data - data intensive computing Performance models for combining data analysis with data movement (between caches, disks, archives)

National Partnership for Advanced Computational Infrastructure Heuristics Experience based models of computer usage Dependent on computer architecture Presence of data caches, memory-mapped I/O Architectures used at SDSC CRAY vector computers X/MP, Y/MP, C-90, T-90 Parallel computers MPPs - Ipsc 860, Paragon, T3D, T3E Clusters - SP

National Partnership for Advanced Computational Infrastructure Supercomputer Data Flow Model CPU Memory Local Disk Archive Disk Archive tape

National Partnership for Advanced Computational Infrastructure Y-MP Heuristics Utilization measured on Cray Y-MP Real memory architecture - entire job context is in memory, no paging of data Exceptional memory bandwidth I/O rate from CPU to memory was 28 Bytes per cycle Maximum execution rate was 2 Flops per cycle Scaled memory on C-90 to test heuristics Noted that increasing memory from 1 GB to 2 GBs decreased idle time from 10% to 2 % Sustained execution rate was 1.8 GFlops

National Partnership for Advanced Computational Infrastructure Data Generation Metrics CPUMemory Local Disk Archive Disk Archive tape 7 Bytes/Flop 1 Byte/60 Flop 1 Byte of storage per Flops 1/7 of data persists for a day 1/7 of data sent to archive Hold data forever Hold data for 1 week Hold data for 1 day All data sent to tape

National Partnership for Advanced Computational Infrastructure Peak Teraflops System Compute Engine Local Disk Archive Disk Archive Tape TB memory Sustain ? GF ? GB/sec ? TB 1 day cache ? MB/sec 1 week cache ? MB/sec ? TB ? PB TeraFlops System

National Partnership for Advanced Computational Infrastructure Data Sizes on Disk How much scratch space is used by each job? Disk space is times the memory size. Data lasts for about one day Average execution time for long running jobs 30 minutes to 1 hour For jobs using all of memory Between 48 and 24 jobs per day Each job uses (Disk space) / (Number of jobs) Or 40/48 Memory = 80% of memory

National Partnership for Advanced Computational Infrastructure Peak Teraflops Data Flow Model Compute Engine Local Disk Archive Disk Archive Tape TB memory Sustain 150 GF 1 GB/sec 10 TB 1 day cache 40 MB/sec 1 week cache 40 MB/sec 5 TB PB TeraFlops System

National Partnership for Advanced Computational Infrastructure HPSS Archival Storage System 108 GB SSA RAID High Performance Gateway Node High Node Disk Mover HiPPI driver Wide Node Disk Mover HiPPI driver 54 GB SSA RAID 108 GB SSA RAID 108 GB SSA RAID 54 GB SSA RAID 108 GB SSA RAID 108 GB SSA RAID Silver Node Storage / Purge Bitfile / Migration Nameservice/PVL Log Daemon Silver Node Tape / disk mover DCE / FTP /HIS Log Client 160 GB SSA RAID Silver Node Tape / disk mover DCE / FTP /HIS Log Client 830 GB MaxStrat RAID 9490 Robot Four Drives 3490 Tape RS6000 Tape Mover PVR (9490) HiPPI Switch Trail- Blazer3 Switch Silver Node Tape / disk mover DCE / FTP /HIS Log Client Silver Node Tape / disk mover DCE / FTP /HIS Log Client Silver Node Tape / disk mover DCE / FTP /HIS Log Client Silver Node Tape / disk mover DCE / FTP /HIS Log Client Silver Node Tape / disk mover DCE / FTP /HIS Log Client Magstar 3590 Tape 3494 Robot Eight Tape Drives Magstar 3590 Tape 3494 Robot Seven Tape Drives

National Partnership for Advanced Computational Infrastructure Equivalent of Ohm’s Law for Computer Science How does one relate application requirements to computation rates and I/O bandwidths? Use prototype data movement problem to derive physical parameters that characterize applications.

National Partnership for Advanced Computational Infrastructure Data Distribution Comparison Data Handling Platform Supercomputer Execution rate r<R Bandwidths linking systems areB & b Operations per bit for analysis is C Operations per bit for data transfer isc Reduce size of data from S bytes to s bytes and analyze Should the data reduction be done before transmission? Data Bb

National Partnership for Advanced Computational Infrastructure Distributing Services Compare times for analyzing data with size reduction from S to s Read Data Reduce Data Transmit Data NetworkReceive Data Read Data Reduce Data Transmit Data Network Receive Data S / BC S / rc s / rs / bc s / R c S / Rc S / rS / bC S / RS / B Data Handling Platform Supercomputer Data Handling Platform Supercomputer

National Partnership for Advanced Computational Infrastructure Comparison of Time T(Super) = S/B + CS/r + cs/r + s/b + cs/R Processing at supercomputer Processing at archive T(Archive) = S/B + cS/r + S/b + cS/R + CS/R

National Partnership for Advanced Computational Infrastructure Optimization Parameter Selection Have algebraic equation with eight independent variables. T (Super) < T (Archive) S/B + CS/r + cs/r + s/b + cs/R < S/B + cS/r + S/b + cS/R + CS/R Which variable provides the simplest optimization criterion?

National Partnership for Advanced Computational Infrastructure Scaling Parameters Data size reduction ratio s/S Execution slow down ratior/R Problem complexityc/C Communication/Execution balancer/(cb) When r/(cb) = 1, the data processing rate is the same as the data transmission rate. Optimal designs have r/(cb) = 1 Note (r/c) is the number of bits/sec that can be processed.

National Partnership for Advanced Computational Infrastructure Bandwidth Optimization Moving all of the data is faster, T(Super) < T(Archive) Sufficiently fast network b > (r /C) (1 - s/S) / [1 - r/R - (c/C) (1 + r/R) (1 - s/S)] Note the denominator changes sign when C < c (1 + r/R) / [(1 - r/R) (1 - s/S)] Even with an infinitely fast network, it is better to do the processing at the archive if the complexity is too small.

National Partnership for Advanced Computational Infrastructure Execution Rate Optimization Moving all of the data is faster, T(Super) < T(Archive) Sufficiently fast supercomputer R > r [1 + (c/C) (1 - s/S)] / [1 - (c/C) (1 - s/S) (1 + r/(cb)] Note the denominator changes sign when C < c (1 - s/S) [1 + r/(cb)] Even with an infinitely fast supercomputer, it is better to process at the archive if the complexity is too small.

National Partnership for Advanced Computational Infrastructure Data Reduction Optimization Moving all of the data is faster, T(Super) < T(Archive) Data reduction is small enough s > S {1 - (C/c)(1 - r/R) / [1 + r/R + r/(cb)]} Note criteria changes sign when C > c [1 + r/R + r/(cb)] / (1 - r/R) When the complexity is sufficiently large, it is faster to process on the supercomputer even when data can be reduced to one bit.

National Partnership for Advanced Computational Infrastructure Complexity Analysis Moving all of the data is faster, T(Super) < T(Archive) Sufficiently complex analysis C > c (1-s/S) [1 + r/R + r/(cb)] / (1-r/R) Note, as the execution ratio approaches 1, the required complexity becomes infinite Also, as the amount of data reduction goes to zero, the required complexity goes to zero.

National Partnership for Advanced Computational Infrastructure Characterization of Supercomputer Systems Sufficiently high complexity Move data to processing engine Digital Library execution of remote services Traditional supercomputer processing of applications Sufficiently low complexity Move process to the data source Metacomputing execution of remote applications Traditional digital library service

National Partnership for Advanced Computational Infrastructure Computer Architectures Processor in memory Do computations within memory Complexity of supported operations Commodity processors L2 caches L3 caches Parallel computers Memory bandwidth between nodes MPP - shared memory Cluster - distributed memory

National Partnership for Advanced Computational Infrastructure Characterization Metric Describe systems in terms of their balance Optimal designs have r/(cb) = 1 Equivalent of Ohm’s law R = C B Characterize applications in terms of their complexity Operations per byte of data C = R / B

National Partnership for Advanced Computational Infrastructure Second Example Inclusion of latency (time for process to start) and overhead (time to execute communication protocol) Illustrate with combined optimization of use of network and CPU

National Partnership for Advanced Computational Infrastructure Optimizing Use of Resources Compare time needed to do calculations with time needed to access data over a network Time spent using a CPU = Execution time + protocol processing time = Cc * Sc / Rc + Cp * St / Rp Where St = size of transmitted data (bytes) Sc = size of application data (bytes) Cc = number of operations per byte of transmitted data for the application Cp = number of operations per byte to process protocol Rc = execution rate of application Rp = execution rate of protocol

National Partnership for Advanced Computational Infrastructure Characterizing Latency Time during which a network transmits data = Latency for initiating transfer + transmission time = L + St / B Where L is the round trip latency at the speed of light (sec) B is the bandwidth (bytes/sec)

National Partnership for Advanced Computational Infrastructure Solve for Balanced System CPU utilization time = Network utilization time Solve for transmission size as a function of Sc/St St = L B / [B * Cp / Rp + (B * Cc / Rc) * (Sc / St) -1] Solution exists when Sc/St > [Rc / (B*Cc)] [1 - B*Cp / Rp] and B * Cp / Rp < 1

National Partnership for Advanced Computational Infrastructure Comparing Utilization of Resources Network utilization Un = Transmission time / (Transmission + latency) = 1 / [1 + (L * B / St)] CPU utilization Uc = Execution time / (Execution + Protocol processing) = 1 / [1 + (Cp * Rc) / (Cc * Rp) * (St / Sc)] Define h = Sc / St

National Partnership for Advanced Computational Infrastructure Comparing Efficiencies h = S-compute / S-transmit Utilization U-cpu U-network

National Partnership for Advanced Computational Infrastructure Crossover Point When utilization of bandwidth and execution resources is balanced: 1 / [1 + (L * B / St)] = 1 / [1 + (Cp * Rc) / (Cc * Rp) / h] For optimal St, solve for h = Sc/St, and find h = (Rc Cp / 2 Rp Cc) [ sqrt(1 + 4 Rp / Cp B) -1] For small B * Cp / Rp h ~ Rc / Cc B or St / B ~ Sc Cc / Rc And transmission time ~ execution time

National Partnership for Advanced Computational Infrastructure Application Summary Optimal application for a given architecture B * Cc / Rc ~ 1 (Bytes/sec) (Operations/byte) / (Operations/sec) Cc ~ Rc / B Also need cost of network utilization to be small B * Cp / Rp < 1 And amount of data transmitted proportional to latency St = L B / [B * Cp / Rp + (B * Cc / Rc) * (Sc / St) -1]

National Partnership for Advanced Computational Infrastructure Further Information