Supercomputing in Plain English: Overview

Supercomputing in Plain English: Overview

Supercomputing in Plain English Distributed Multiprocessing Henry Neeman, University of Oklahoma Director, OU Supercomputing Center for Education & Research (OSCER) Assistant Vice President, Information Technology Research Strategy Advisor Associate Professor, Gallogly College of Engineering Adjunct Associate Professor, School of Computer Science Tuesday March 6 2018 This is an experiment! Its the nature of these kinds of videoconferences that FAILURES ARE GUARANTEED TO HAPPEN! NO PROMISES! So, please bear with us. Hopefully everything will work out well enough. If you lose your connection, you can retry the same kind of connection, or try connecting another way. Remember, if all else fails, you always have the phone bridge

to fall back on. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 2 PLEASE MUTE YOURSELF No matter how you connect, PLEASE MUTE YOURSELF, so that we cannot hear you. At OU, we will turn off the sound on all conferencing technologies. That way, we wont have problems with echo cancellation. Of course, that means we cannot hear questions. So for questions, youll need to send e-mail: [email protected] PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF.

Supercomputing in Plain English: Distributed Par Tue March 6 2018 3 Download the Slides Beforehand Before the start of the session, please download the slides from the Supercomputing in Plain English website: http://www.oscer.ou.edu/education/ That way, if anything goes wrong, you can still follow along with just audio. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 4

Zoom Go to: http://zoom.us/j/979158478 Many thanks Eddie Huebsch, OU CIO, for providing this. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 5 YouTube You can watch from a Windows, MacOS or Linux laptop or an Android or iOS handheld using YouTube. Go to YouTube via your preferred web browser or app, and then search for: Supercomputing InPlainEnglish (InPlainEnglish is all one word.) Many thanks to Skyler Donahue of OneNet for providing this.

PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 6 Twitch You can watch from a Windows, MacOS or Linux laptop or an Android or iOS handheld using Twitch. Go to: http://www.twitch.tv/sipe2018 Many thanks to Skyler Donahue of OneNet for providing this. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018

7 Wowza #1 You can watch from a Windows, MacOS or Linux laptop using Wowza from the following URL: http://jwplayer.onenet.net/streams/sipe.html If that URL fails, then go to: http://jwplayer.onenet.net/streams/sipebackup.html Many thanks to Skyler Donahue of OneNet for providing this. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 8 Wowza #2 Wowza has been tested on multiple browsers on each of:

Windows 10: IE, Firefox, Chrome, Opera, Safari MacOS: Safari, Firefox Linux: Firefox, Opera Weve also successfully tested it via apps on devices with: Android iOS Many thanks to Skyler Donahue of OneNet for providing this. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 9 Toll Free Phone Bridge IF ALL ELSE FAILS, you can use our US TOLL phone bridge: 405-325-6688 684 684 # NOTE: This is for US call-ins ONLY. PLEASE MUTE YOURSELF and use the phone to listen.

Dont worry, well call out slide numbers as we go. Please use the phone bridge ONLY IF you cannot connect any other way: the phone bridge can handle only 100 simultaneous connections, and we have over 1000 participants. Many thanks to OU CIO Eddie Huebsch for providing the phone bridge.. Supercomputing in Plain English: Distributed Par Tue March 6 2018 10 Please Mute Yourself No matter how you connect, PLEASE MUTE YOURSELF, so that we cannot hear you. (For YouTube, Twitch and Wowza, you dont need to do that, because the information only goes from us to you, not from you to us.) At OU, we will turn off the sound on all conferencing technologies. That way, we wont have problems with echo cancellation. Of course, that means we cannot hear questions.

So for questions, youll need to send e-mail. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 11 Questions via E-mail Only Ask questions by sending e-mail to: [email protected] All questions will be read out loud and then answered out loud. DONT USE CHAT OR VOICE FOR QUESTIONS! No one will be monitoring any of the chats, and if we can hear your question, youre creating an echo cancellation problem. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 12

Onsite: Talent Release Form If youre attending onsite, you MUST do one of the following: complete and sign the Talent Release Form, OR sit behind the cameras (where you cant be seen) and dont talk at all. If you arent onsite, then PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 13 TENTATIVE Schedule Tue Jan 23: Storage: What the Heck is Supercomputing? Tue Jan 30: The Tyranny of the Storage Hierarchy Part I Tue Feb 6: The Tyranny of the Storage Hierarchy Part II Tue Feb 13: Instruction Level Parallelism Tue Feb 20: Stupid Compiler Tricks Tue Feb 27: Distributed Par Multithreading Tue March 6: Distributed Multiprocessing

Tue March 13: NO SESSION (Henry business travel) Tue March 20: NO SESSION (OU's Spring Break) Tue March 27: Applications and Types of Parallelism Tue Apr 3: Multicore Madness Tue Apr 10: High Throughput Computing Tue Apr 17: NO SESSION (Henry business travel) Tue Apr 24: GPGPU: Number Crunching in Your Graphics Card Tue May 1: Grab Bag: Scientific Libraries, I/O Libraries, Visualization Supercomputing in Plain English: Distributed Par Tue March 6 2018 14 Thanks for helping! OU IT

OSCER operations staff (Dave Akin, Patrick Calhoun, Kali McLennan, Jason Speckman, Brett Zimmerman) OSCER Research Computing Facilitators (Jim Ferguson, Horst Severini) Debi Gentis, OSCER Coordinator Kyle Dudgeon, OSCER Manager of Operations Ashish Pai, Managing Director for Research IT Services The OU IT network team OU CIO Eddie Huebsch OneNet: Skyler Donahue Oklahoma State U: Dana Brunson

Supercomputing in Plain English: Distributed Par Tue March 6 2018 15 This is an experiment! Its the nature of these kinds of videoconferences that FAILURES ARE GUARANTEED TO HAPPEN! NO PROMISES! So, please bear with us. Hopefully everything will work out well enough. If you lose your connection, you can retry the same kind of connection, or try connecting another way. Remember, if all else fails, you always have the phone bridge to fall back on. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018

16 Coming in 2018! Coalition for Advancing Digital Research & Education (CADRE) Conference: 17-18 2018 @ Oklahoma State U, Stillwater OK USA Apr https://hpcc.okstate.edu/cadre-conference Linux Clusters Institute workshops http://www.linuxclustersinstitute.org/workshops/ Introductory HPC Cluster System Administration: May 14-18 2018 @ U Nebraska, Lincoln NE USA Intermediate HPC Cluster System Administration: Aug 13-17 2018 @ Yale U, New Haven CT USA Great Plains Network Annual Meeting: details coming soon Advanced Cyberinfrastructure Research & Education Facilitators (ACI-REF) Virtual Residency Aug 5-10 2018, U Oklahoma, Norman OK USA PEARC 2018, July 22-27, Pittsburgh PA USA https://www.pearc18.pearc.org/

IEEE Cluster 2018, Sep 10-13, Belfast UK https://cluster2018.github.io OKLAHOMA SUPERCOMPUTING SYMPOSIUM 2018, Sep 25-26 2018 @ OU SC18 supercomputing conference, Nov 11-16 2018, Dallas TX USA http://sc18.supercomputing.org/ Supercomputing in Plain English: Distributed Par Tue March 6 2018 17 Outline The Desert Islands Analogy Distributed Parallelism MPI

Supercomputing in Plain English: Distributed Par Tue March 6 2018 18 The Desert Islands Analogy An Island Hut Imagine youre on an island in a little hut. Inside the hut is a desk. On the desk is:

a phone; a pencil; a calculator; a piece of paper with instructions; a piece of paper with numbers (data). Instructions: What to Do ... Add the number in slot 27 to the number in slot 239, and put the result in slot 71. if the number in slot 71 is equal to the number in slot 118 then Call 555-0127 and leave a voicemail containing the number in slot 962. else Call your voicemail box and collect a voicemail from 555-0063, and put that number in slot 715. DATA 1.

2. 3. 4. 5. 6. 7. 8. 9. ... 27.3 -491.41 24 -1e-05 141.41 0 4167 94.14 -518.481 ...

Supercomputing in Plain English: Distributed Par Tue March 6 2018 20 Instructions The instructions are split into two kinds: Arithmetic/Logical for example: Add the number in slot 27 to the number in slot 239, and put the result in slot 71. Compare the number in slot 71 to the number in slot 118, to see whether they are equal. Communication for example: Call 555-0127 and leave a voicemail containing the number in slot 962. Call your voicemail box and collect a voicemail from 555-0063, and put that number in slot 715. Supercomputing in Plain English: Distributed Par

Tue March 6 2018 21 Is There Anybody Out There? If youre in a hut on an island, you arent specifically aware of anyone else. Especially, you dont know whether anyone else is working on the same problem as you are, and you dont know whos at the other end of the phone line. All you know is what to do with the voicemails you get, and what phone numbers to send voicemails to. Supercomputing in Plain English: Distributed Par Tue March 6 2018 22 Someone Might Be Out There Now suppose that Horst is on another island somewhere, in the same kind of hut, with the same kind of equipment.

Suppose that he has the same list of instructions as you, but a different set of numbers (both data and phone numbers). Like you, he doesnt know whether theres anyone else working on his problem. Supercomputing in Plain English: Distributed Par Tue March 6 2018 23 Even More People Out There Now suppose that Bruce and Dee are also in huts on islands. Suppose that each of the four has the exact same list of instructions, but different lists of numbers. And suppose that the phone numbers that people call are each others: that is, your instructions have you call Horst, Bruce and Dee, Horsts has him call Bruce, Dee and you, and so on. Then you might all be working together on the same problem. Supercomputing in Plain English: Distributed Par Tue March 6 2018

24 All Data Are Private Notice that you cant see Horsts or Bruces or Dees numbers, nor can they see yours or each others. Thus, everyones numbers are private: theres no way for anyone to share numbers, except by leaving them in voicemails. Supercomputing in Plain English: Distributed Par Tue March 6 2018 25 Long Distance Calls: 2 Costs When you make a long distance phone call, you typically have to pay two costs: Connection charge: the fixed cost of connecting your phone to someone elses, even if youre only connected for a second Per-minute charge: the cost per minute of talking, once

youre connected If the connection charge is large, then you want to make as few calls as possible. See: http://www.youtube.com/watch?v=8k1UOEYIQRo Supercomputing in Plain English: Distributed Par Tue March 6 2018 26 Distributed Parallelism Like Desert Islands Distributed parallelism is very much like the Desert Islands analogy: processes are independent of each other. All data are private. Processes communicate by passing messages (like voicemails).

The cost of passing a message is split into: latency (connection time) bandwidth (time per byte) Supercomputing in Plain English: Distributed Par Tue March 6 2018 28 Latency vs Bandwidth on Schooner In 2018, a benchmark of the Infiniband interconnect on the University of Oklahomas Linux cluster revealed: Latency the time for the first bit to show up at the destination is ~1.26 microseconds; Bandwidth the speed of the subsequent bits is ~37.2 Gigabits per second (~0.027 nanosec per bit). Thus, on OUs cluster Infiniband:

the first bit of a message shows up in ~1260 nanosec; the last bit of a message shows up in ~0.027 nanosec. So latency is ~47,000 times worse than bandwidth! Supercomputing in Plain English: Distributed Par Tue March 6 2018 29 Latency vs Bandwidth on Schooner In 2018, a benchmark of the Infiniband interconnect on the University of Oklahomas Linux cluster revealed: Latency the time for the first bit to show up at the destination is ~1.26 microseconds; Bandwidth the speed of the subsequent bits is ~37.2 Gigabits per second (~0.027 nanosec per bit). Thus, on OUs cluster Infiniband: the first bit of a message shows up in ~1260 nanosec; the last bit of a message shows up in ~0.027 nanosec. So latency is ~47,000 times worse than bandwidth! Thats like having a long distance service that charges:

$470 to make a call at all, regardless of duration; 1 per minute after the first 33 days on the call. Supercomputing in Plain English: Distributed Par Tue March 6 2018 30 MPI: The Message-Passing Interface Most of this discussion is from [1] and [2]. What Is MPI? The Message-Passing Interface (MPI) is a standard for expressing distributed parallelism via message passing. MPI consists of a header file, a library of routines and a runtime environment. When you compile a program that has MPI calls in it, your compiler links to a local implementation of MPI, and then you get parallelism; if the MPI library isnt available,

then the compile will fail. MPI can be used in Fortran, C and C++. There are also unofficial bindings for MATLAB, Python, R and a few others, but these arent part of the official MPI standard. Supercomputing in Plain English: Distributed Par Tue March 6 2018 32 MPI Calls In C, MPI calls look like: mpi_error_code = MPI_Funcname(); In Fortran, MPI calls look like this: CALL MPI_Funcname(, mpi_error_code) Notice that mpi_error_code is returned by the MPI routine MPI_Funcname, with a value of MPI_SUCCESS indicating that MPI_Funcname has worked correctly. In C++, MPI calls look like: mpi_error_code = MPI::Funcname(); But, the C++ binding has been deprecated, so DONT USE IT. Instead, use the C binding, above.

Supercomputing in Plain English: Distributed Par Tue March 6 2018 33 MPI is an API MPI is actually just an Application Programming Interface (API). An API specifies what a call to each routine should look like, and how each routine should behave. An API does not specify how each routine should be implemented, and sometimes is intentionally vague about certain aspects of a routines behavior. Each platform can have its own MPI implementation or multiple MPI implementations. Supercomputing in Plain English: Distributed Par Tue March 6 2018 34 Example MPI Implementations

MPICH2 (http://www.mpich.org) OpenMPI (https://www.open-mpi.org) Intel MPI (https://software.intel.com/en-us/intel-mpi-library) Microsoft MPI ( https://msdn.microsoft.com/en-us/library/bb524831(v=vs.85).aspx) IBM Platform MPI ( https://www.ibm.com/support/knowledgecenter/en/SSF4ZA_9.1.3/pmpi _welcome/pmpi_9.1.3.html ) IBM Parallel Operating Environment (

https://www.ibm.com/support/knowledgecenter/SSFK3V_2.3.0/com.ibm .cluster.pe.v2r3.pe400.doc/am106_mpibeo.htm) Cray Message Passing Toolkit ( https://pubs.cray.com/content/S-2529/17.05/xctm-series-programming-e nvironment-user-guide-1705-s-2529/mpt Supercomputing in Plain English: Distributed Par ) Tue March 6 2018 35 WARNING! In principle, the MPI standard provides bindings for: C C++ (deprecated) Fortran 77 Fortran 90 In practice, you should do this: To use MPI in a C++ code, use the C binding. To use MPI in Fortran 90, use the Fortran 77 binding. This is because the C++ and Fortran 90 bindings are

less popular, and therefore less well tested. Supercomputing in Plain English: Distributed Par Tue March 6 2018 36 The 6 Most Important MPI Routines MPI_Init starts up the MPI runtime environment at the beginning of a run.

MPI_Finalize shuts down the MPI runtime environment at the end of a run. MPI_Comm_size gets the number of processes in a run, Np (typically called just after MPI_Init). MPI_Comm_rank gets the process ID that the current process uses, which is between 0 and Np-1 inclusive (typically called just after MPI_Init). MPI_Send sends a message from the current process to some other process (the destination). MPI_Recv receives a message on the current process from some other process (the source). Supercomputing in Plain English: Distributed Par Tue March 6 2018 37 More Example MPI Routines MPI_Bcast broadcasts a message from one process to all of the others. MPI_Reduce performs a reduction (for example, sum, maximum) of a variable on all processes, sending the result to a

single process. NOTE: Here, reduce means turn many values into fewer values. MPI_Gather gathers a set of subarrays, one subarray from each process, into a single large array on a single process. MPI_Scatter scatters a single large array on a single process into subarrays, one subarray sent to each process. Routines that use all processes at once are known as collective; Supercomputing Distributed Par routines that involve only ain Plain fewEnglish: are known as point-to-point. Tue March 6 2018 38

MPI Program Structure (C) #include #include #include [other includes] int main (int argc, char* argv[]) { /* main */ int my_rank, num_procs, mpi_error_code; [other declarations] mpi_error_code = MPI_Init(&argc, &argv); /* Start up MPI */ mpi_error_code = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); mpi_error_code = MPI_Comm_size(MPI_COMM_WORLD, &num_procs); [actual work goes here] mpi_error_code = MPI_Finalize(); /* Shut down MPI */ } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018

39 MPI is SPMD MPI uses kind of parallelism known as Single Program, Multiple Data (SPMD). This means that you have one MPI program a single executable that is executed by all of the processes in an MPI run. So, to differentiate the roles of various processes in the MPI run, you have to have if statements: if (my_rank == server_rank) { } Supercomputing in Plain English: Distributed Par Tue March 6 2018 40

Example: Hello World 1. Start the MPI system. 2. Get this processs rank, and the number of processes. 3. Output Hello world along with the rank and number of processes. 4. Shut down the MPI system. Supercomputing in Plain English: Distributed Par Tue March 6 2018 41 Example: Hello World Code (C) #include #include #include "mpi.h" int main (int argc, char** argv) { /* main */ int number_of_processes; int my_rank; int mpi_error_code;

mpi_error_code = MPI_Init(&argc, &argv); mpi_error_code = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); mpi_error_code = MPI_Comm_size(MPI_COMM_WORLD, &number_of_processes); printf("%d of %d: Hello, world!\n", my_rank, number_of_processes); mpi_error_code = MPI_Finalize(); } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018 42 Example: Hello World Code (F90) PROGRAM hello_world_mpi IMPLICIT NONE INCLUDE "mpif.h" INTEGER :: number_of_processes, my_rank INTEGER :: mpi_error_code CALL MPI_Init(mpi_error_code) CALL MPI_Comm_rank(MPI_COMM_WORLD, number_of_processes, & &

mpi_error_code) CALL MPI_Comm_size(MPI_COMM_WORLD, my_rank, & & mpi_error_code) PRINT *, my_rank, " of ", number_of_processes, & & ": Hello, world!" CALL MPI_Finalize(mpi_error_code) END PROGRAM hello_world_mpi Supercomputing in Plain English: Distributed Par Tue March 6 2018 43 Example: Hello World Output 2 of 20: Hello, world! 4 of 20: Hello, world! 8 of 20: Hello, world!

10 of 20: Hello, world! 14 of 20: Hello, world! 15 of 20: Hello, world! 16 of 20: Hello, world! 17 of 20: Hello, world! 18 of 20: Hello, world! 0 of 20: Hello, world! 1 of 20: Hello, world! 3 of 20: Hello, world! 5 of 20: Hello, world! 6 of 20: Hello, world! 7 of 20: Hello, world! 9 of 20: Hello, world! 11 of 20: Hello, world! 12 of 20: Hello, world! 13 of 20: Hello, world! 19 of 20: Hello, world! Supercomputing in Plain English: Distributed Par Tue March 6 2018

44 Example: Greetings 1. 2. 3. Start the MPI system. Get this processs rank, and the number of processes. If Im not the server process: 1. 2. 4. Create a greeting string. Send it to the server process. If I am the server process: 1.

For each of the client processes: 1. 2. Receive its greeting string. Print its greeting string. 5. Shut down the MPI system. See [1]. Supercomputing in Plain English: Distributed Par Tue March 6 2018 45 greeting.c #include #include #include int main (int argc, char* argv[]) { /* main */

const int maximum_message_length = 100; const int server_rank = 0; char message[maximum_message_length+1]; MPI_Status status; /* Info about receive status int my_rank; /* This process ID int num_procs; /* Number of processes in run int source; /* Process ID to receive from int destination; /* Process ID to send to int

tag = 0; /* Message ID int mpi_error_code; /* Error code for MPI calls [work goes here] } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018 */ */ */ */ */ */ */ 46 Greetings Startup/Shutdown

[header file includes] int main (int argc, char* argv[]) { /* main */ [declarations] mpi_error_code = MPI_Init(&argc, &argv); mpi_error_code = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); mpi_error_code = MPI_Comm_size(MPI_COMM_WORLD, &num_procs); if (my_rank != server_rank) { [work of each non-server (worker) process] } /* if (my_rank != server_rank) */ else { [work of server process] } /* if (my_rank != server_rank)else */ mpi_error_code = MPI_Finalize(); } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018 47

Greetings Clients Work [header file includes] int main (int argc, char* argv[]) { /* main */ [declarations] [MPI startup (MPI_Init etc)] if (my_rank != server_rank) { sprintf(message, "Greetings from process #%d!", my_rank); destination = server_rank; mpi_error_code = MPI_Send(message, strlen(message) + 1, MPI_CHAR, destination, tag, MPI_COMM_WORLD); } /* if (my_rank != server_rank) */ else { [work of server process] } /* if (my_rank != server_rank)else */ mpi_error_code = MPI_Finalize(); } /* main */ Supercomputing in Plain English: Distributed Par

Tue March 6 2018 48 Greetings Servers Work [header file includes] int main (int argc, char* argv[]) { /* main */ [declarations, MPI startup] if (my_rank != server_rank) { [work of each client process] } /* if (my_rank != server_rank) */ else { for (source = 0; source < num_procs; source++) { if (source != server_rank) { mpi_error_code = MPI_Recv(message, maximum_message_length + 1, MPI_CHAR, source, tag, MPI_COMM_WORLD, &status); fprintf(stderr, "%s\n", message); } /* if (source != server_rank) */

} /* for source */ } /* if (my_rank != server_rank)else */ mpi_error_code = MPI_Finalize(); } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018 49 How an MPI Run Works Every process gets a copy of the executable: Single Program, Multiple Data (SPMD). They all start executing it. Each looks at its own rank to determine which part of the

problem to work on. Each process works completely independently of the other processes, except when communicating. Supercomputing in Plain English: Distributed Par Tue March 6 2018 50 Compiling and Running % mpicc -o greeting_mpi greeting.c % mpirun -np 1 greeting_mpi % mpirun -np 2 greeting_mpi Greetings from process #1! % mpirun -np 3 greeting_mpi Greetings from process #1! Greetings from process #2! % mpirun Greetings Greetings Greetings

-np from from from 4 greeting_mpi process #1! process #2! process #3! Note: The compile command and the run command vary from platform to platform. This ISNT how you run MPI on Schooner. Supercomputing in Plain English: Distributed Par Tue March 6 2018 51 Why is Rank #0 the Server? const int server_rank = 0;

By convention, if an MPI program uses a client-server approach, then the server process has rank (process ID) #0. Why? A run must use at least one process but can use multiple processes. Process ranks are 0 through Np-1, for Np >1 , where Np is the number of processes in the run. Therefore, every MPI run has a process with rank #0. Note: Every MPI run also has a process with rank Np-1, so you could use Np-1 as the server instead of 0 but no one does. Supercomputing in Plain English: Distributed Par Tue March 6 2018 52 Does There Have to be a Server? There DOESNT have to be a server. Its perfectly possible to write an MPI code that has no server as such. For example, weather forecasting and other transport codes typically share most duties equally, and likewise chemistry

and astronomy codes. In practice, though, most codes use rank #0 to do things like small scale I/O, since its typically more efficient to have one process read small files and then broadcast small input data to the other processes, or to gather the output data and write it to disk. Supercomputing in Plain English: Distributed Par Tue March 6 2018 53 Why Rank? Why does MPI use the term rank to refer to process ID? In general, a process has an identifier that is assigned by the operating system (for example, Unix), and that is unrelated to MPI: % ps PID TTY TIME CMD 52170812 ttyq57 0:01 tcsh Also, each processor has an identifier, but an MPI run that

uses fewer than all processors will use an arbitrary subset. The rank of an MPI process is neither of these. Supercomputing in Plain English: Distributed Par Tue March 6 2018 54 Compiling and Running Recall: % mpicc -o greeting_mpi greeting.c % mpirun -np 1 greeting_mpi % mpirun -np 2 greeting_mpi Greetings from process #1! % mpirun -np 3 greeting_mpi Greetings from process #1! Greetings from process #2! % mpirun Greetings Greetings Greetings

-np from from from 4 greeting_mpi process #1! process #2! process #3! Supercomputing in Plain English: Distributed Par Tue March 6 2018 55 Deterministic Operation? % mpirun Greetings Greetings Greetings

-np from from from 4 greeting_mpi process #1! process #2! process #3! The order in which the greetings are output is deterministic. Why? for (source = 0; source < num_procs; source++) { if (source != server_rank) { mpi_error_code = MPI_Recv(message, maximum_message_length + 1, MPI_CHAR, source, tag, MPI_COMM_WORLD, &status); fprintf(stderr, "%s\n", message); } /* if (source != server_rank) */

} /* for source */ This loop ignores the order in which messages are received . Supercomputing in Plain English: Distributed Par Tue March 6 2018 56 Deterministic Parallelism for (source = 0; source < num_procs; source++) { if (source != server_rank) { mpi_error_code = MPI_Recv(message, maximum_message_length + 1, MPI_CHAR, source, tag, MPI_COMM_WORLD, &status); fprintf(stderr, "%s\n", message); } /* if (source != server_rank) */ } /* for source */ Because of the order in which the loop iterations occur, the

greeting messages will be output in deterministic order, regardless of the order in which the greeting messages are received. In principle, the run could pause for a long time, waiting for one client processs message to arrive at the server process. Supercomputing in Plain English: Distributed Par Tue March 6 2018 57 Nondeterministic Parallelism for (source = 0; source < num_procs; source++) { if (source != server_rank) { mpi_error_code = MPI_Recv(message, maximum_message_length + 1, MPI_CHAR, MPI_ANY_SOURCE, tag, MPI_COMM_WORLD, &status); fprintf(stderr, "%s\n", message); } /* if (source != server_rank) */ } /* for source */

Because of this change, the greeting messages will be output in non-deterministic order, specifically in the order in which theyre received. Supercomputing in Plain English: Distributed Par Tue March 6 2018 58 Message = Envelope + Contents MPI_Send(message, strlen(message) + 1, MPI_CHAR, destination, tag, MPI_COMM_WORLD); When MPI sends a message, it doesnt just send the contents; it also sends an envelope describing the contents: Size (number of elements of the messages data type) Data type Source: rank of sending process Destination: rank of process to receive

Tag (message ID) Communicator (for example, MPI_COMM_WORLD) Supercomputing in Plain English: Distributed Par Tue March 6 2018 59 MPI Data Types C Fortran char MPI_CHAR CHARACTER MPI_CHARACTER

int MPI_INT INTEGER MPI_INTEGER float MPI_FLOAT REAL MPI_REAL double MPI_DOUBLE DOUBLE

PRECISION MPI_DOUBLE_PRECISION MPI supports several other data types, but most are variations on these, and probably these are all youll use. Supercomputing in Plain English: Distributed Par Tue March 6 2018 60 Message Tags My daughter was born in mid-December. So, if I give her a present in December, how does she know which of these its for? Her birthday Christmas Hanukkah She knows because of the tag on the present: A little cake with candles means birthday

A little tree or a Santa means Christmas A little menorah means Hanukkah Supercomputing in Plain English: Distributed Par Tue March 6 2018 61 Message Tags for (source = 0; source < num_procs; source++) { if (source != server_rank) { mpi_error_code = MPI_Recv(message, maximum_message_length + 1, MPI_CHAR, source, tag, MPI_COMM_WORLD, &status); fprintf(stderr, "%s\n", message); } /* if (source != server_rank) */ } /* for source */ The greetings are output in deterministic order, not because messages are sent and received in order, but because each has a tag (message identifier), and

MPI_Recv asks for a specific message (by tag) from a specific source (by rank). Supercomputing in Plain English: Distributed Par Tue March 6 2018 62 Parallelism is Nondeterministic for (source = 0; source < num_procs; source++) { if (source != server_rank) { mpi_error_code = MPI_Recv(message, maximum_message_length + 1, MPI_CHAR, MPI_ANY_SOURCE, tag, MPI_COMM_WORLD, &status); fprintf(stderr, "%s\n", message); } /* if (source != server_rank) */ } /* for source */ But here the greetings are output in non-deterministic order. Supercomputing in Plain English: Distributed Par

Tue March 6 2018 63 Communicators An MPI communicator is a collection of processes that can send messages to each other. MPI_COMM_WORLD is the default communicator; it contains all of the processes in the current run. Its probably the only one youll need in most cases. Some libraries create special library-only communicators, which can simplify keeping track of message tags. Supercomputing in Plain English: Distributed Par Tue March 6 2018 64 Broadcasting What happens if one process has data that everyone else needs to know?

For example, what if the server process needs to send an input value to the others? mpi_error_code = MPI_Bcast(&length, 1, MPI_INTEGER, source, MPI_COMM_WORLD); Note that MPI_Bcast doesnt use a tag, and that the call is the same for both the sender and all of the receivers. This is COUNTERINTUITIVE! All processes have to call MPI_Bcast at the same time; everyone waits until everyone is done (synchronization). Supercomputing in Plain English: Distributed Par Tue March 6 2018 65 Broadcast Example: Setup #include #include #include int main (int argc, char** argv) { /* main */

const int server = 0; const int source = server; float* array = (float*)NULL; int length; int num_procs, my_rank, mpi_error_code; mpi_error_code = MPI_Init(&argc, &argv); mpi_error_code = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); mpi_error_code = MPI_Comm_size(MPI_COMM_WORLD, &num_procs); [input, allocate, initialize on server only] [broadcast, output on all processes] mpi_error_code = MPI_Finalize(); } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018 66 Broadcast Example: Input #include #include

#include int main (int argc, char** argv) { /* main */ const int server = 0; const int source = server; float* array = (float*)NULL; int length; int num_procs, my_rank, mpi_error_code; [MPI startup] if (my_rank == server) { scanf("%d", &length); array = (float*)malloc(sizeof(float) * length); for (index = 0; index < length; index++) { array[index] = 0.0; } /* for index */ } /* if (my_rank == server) */ [broadcast , output on all processes] [MPI shutdown] } /* main */ Supercomputing in Plain English: Distributed Par

Tue March 6 2018 67 Broadcast Example: Broadcast #include #include #include int main (int argc, char** argv) { /* main */ const int server = 0; const int source = server; float* array = (float*)NULL; int length; int num_procs, my_rank, mpi_error_code; [MPI startup] [input, allocate, initialize on server only] if (num_procs > 1) { mpi_error_code = MPI_Bcast(&length, 1, MPI_INTEGER, source, MPI_COMM_WORLD); if (my_rank != server) {

array = (float*)malloc(sizeof(float) * length); } /* if (my_rank != server) */ mpi_error_code = MPI_Bcast(array, length, MPI_INTEGER, source, MPI_COMM_WORLD); printf("%d: broadcast length = %d\n", my_rank, length); } /* if (num_procs > 1) */ mpi_error_code = MPI_Finalize(); } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018 68 Broadcast Compile & Run % mpicc -o broadcast broadcast.c % mpirun -np 4 broadcast 0 : broadcast length = 16777216 1 : broadcast length = 16777216 2 : broadcast length = 16777216 3 : broadcast length = 16777216

Supercomputing in Plain English: Distributed Par Tue March 6 2018 69 Reductions A reduction converts an array to a scalar (or, more generally, converts many values to fewer values). For example, sum, product, minimum value, maximum value, Boolean AND, Boolean OR, etc. Reductions are so common, and so important, that MPI has two routines to handle them: MPI_Reduce: sends result to a single specified process MPI_Allreduce: sends result to all processes (and therefore takes longer) Supercomputing in Plain English: Distributed Par Tue March 6 2018 70

Reduction Example #include #include #include int main (int argc, char **argv) { /* main */ const int server = 0; float value, value_sum; int num_procs, my_rank, mpi_error_code; mpi_error_code = MPI_Init(&argc, &argv); mpi_error_code = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); mpi_error_code = MPI_Comm_size(MPI_COMM_WORLD, &num_procs); value_sum = 0.0; value = my_rank * num_procs; mpi_error_code = MPI_Reduce (&value, &value_sum, 1, MPI_FLOAT, MPI_SUM, server, MPI_COMM_WORLD); printf("%d: reduce

value_sum = %d\n", my_rank, value_sum); mpi_error_code = MPI_Allreduce(&value, &value_sum, 1, MPI_FLOAT, MPI_SUM, MPI_COMM_WORLD); printf("%d: allreduce value_sum = %d\n", my_rank, value_sum); mpi_error_code = MPI_Finalize(); } /* main */ Supercomputing in Plain English: Distributed Par Tue March 6 2018 71 Compiling and Running % mpicc -o reduce reduce.c % mpirun -np 4 reduce 3: reduce value_sum = 0 1: reduce value_sum = 0 0: reduce value_sum = 24

2: reduce value_sum = 0 0: allreduce value_sum = 24 1: allreduce value_sum = 24 2: allreduce value_sum = 24 3: allreduce value_sum = 24 Supercomputing in Plain English: Distributed Par Tue March 6 2018 72 Why Two Reduction Routines? MPI has two reduction routines because of the high cost of each communication. If only one process needs the result, then it doesnt make sense to pay the cost of sending the result to all processes. But if all processes need the result, then it may be cheaper to reduce to all processes than to reduce to a single process and then broadcast to all.

Supercomputing in Plain English: Distributed Par Tue March 6 2018 73 Non-blocking Communication MPI allows a process to start a send, then go on and do work while the message is in transit. This is called non-blocking or immediate communication. Here, immediate refers to the fact that the call to the MPI routine returns immediately rather than waiting for the communication to complete. Supercomputing in Plain English: Distributed Par Tue March 6 2018 74 Immediate Send mpi_error_code = MPI_Isend(array, size, MPI_FLOAT,

destination, tag, communicator, &request); Likewise: mpi_error_code = MPI_Irecv(array, size, MPI_FLOAT, source, tag, communicator, &request); This call starts the send/receive, but the send/receive wont be complete until: MPI_Wait(request, status); Whats the advantage of this? Supercomputing in Plain English: Distributed Par Tue March 6 2018 75 Communication Hiding In between the call to MPI_Isend/Irecv and the call to MPI_Wait, both processes can do work!

If that work takes at least as much time as the communication, then the cost of the communication is effectively zero, since the communication wont affect how much work gets done. This is called communication hiding. Supercomputing in Plain English: Distributed Par Tue March 6 2018 76 Rule of Thumb for Hiding When you want to hide communication: as soon as you calculate the data, send it; dont receive it until you need it. That way, the communication has the maximal amount of time to happen in background (behind the scenes). Supercomputing in Plain English: Distributed Par Tue March 6 2018 77

TENTATIVE Schedule Tue Jan 23: Storage: What the Heck is Supercomputing? Tue Jan 30: The Tyranny of the Storage Hierarchy Part I Tue Feb 6: The Tyranny of the Storage Hierarchy Part II Tue Feb 13: Instruction Level Parallelism Tue Feb 20: Stupid Compiler Tricks Tue Feb 27: Distributed Par Multithreading Tue March 6: Distributed Multiprocessing Tue March 13: NO SESSION (Henry business travel) Tue March 20: NO SESSION (OU's Spring Break) Tue March 27: Applications and Types of Parallelism Tue Apr 3: Multicore Madness Tue Apr 10: High Throughput Computing Tue Apr 17: NO SESSION (Henry business travel) Tue Apr 24: GPGPU: Number Crunching in Your Graphics Card Tue May 1: Grab Bag: Scientific Libraries, I/O Libraries, Visualization Supercomputing in Plain English: Distributed Par Tue March 6 2018 78

Thanks for helping! OU IT OSCER operations staff (Dave Akin, Patrick Calhoun, Kali McLennan, Jason Speckman, Brett Zimmerman) OSCER Research Computing Facilitators (Jim Ferguson,

Horst Severini) Debi Gentis, OSCER Coordinator Kyle Dudgeon, OSCER Manager of Operations Ashish Pai, Managing Director for Research IT Services The OU IT network team OU CIO Eddie Huebsch OneNet: Skyler Donahue Oklahoma State U: Dana Brunson Supercomputing in Plain English: Distributed Par Tue March 6 2018 79 This is an experiment! Its the nature of these kinds of videoconferences that FAILURES ARE GUARANTEED TO HAPPEN! NO PROMISES! So, please bear with us. Hopefully everything will work out well enough.

If you lose your connection, you can retry the same kind of connection, or try connecting another way. Remember, if all else fails, you always have the phone bridge to fall back on. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. PLEASE MUTE YOURSELF. Supercomputing in Plain English: Distributed Par Tue March 6 2018 80 Coming in 2018! Coalition for Advancing Digital Research & Education (CADRE) Conference: 17-18 2018 @ Oklahoma State U, Stillwater OK USA Apr https://hpcc.okstate.edu/cadre-conference Linux Clusters Institute workshops

http://www.linuxclustersinstitute.org/workshops/ Introductory HPC Cluster System Administration: May 14-18 2018 @ U Nebraska, Lincoln NE USA Intermediate HPC Cluster System Administration: Aug 13-17 2018 @ Yale U, New Haven CT USA Great Plains Network Annual Meeting: details coming soon Advanced Cyberinfrastructure Research & Education Facilitators (ACI-REF) Virtual Residency Aug 5-10 2018, U Oklahoma, Norman OK USA PEARC 2018, July 22-27, Pittsburgh PA USA https://www.pearc18.pearc.org/ IEEE Cluster 2018, Sep 10-13, Belfast UK https://cluster2018.github.io OKLAHOMA SUPERCOMPUTING SYMPOSIUM 2018, Sep 25-26 2018 @ OU SC18 supercomputing conference, Nov 11-16 2018, Dallas TX USA http://sc18.supercomputing.org/ Supercomputing in Plain English: Distributed Par Tue March 6 2018 81

Thanks for your attention! Questions? www.oscer.ou.edu References [1] P.S. Pacheco, Parallel Programming with MPI, Morgan Kaufmann Publishers, 1997. [2] W. Gropp, E. Lusk and A. Skjellum, Using MPI: Portable Parallel Programming with the Message-Passing Interface, 2nd ed. MIT Press, 1999. Supercomputing in Plain English: Distributed Par Tue March 6 2018 83

Recently Viewed Presentations

  • Descrição E Análise Morfossintática Do Inglês Class 7 ...

    Descrição E Análise Morfossintática Do Inglês Class 7 ...

    2. Words are made up of morphemes, the smallest grammatical units. Morphemes are abstract grammatical entities that are realised as morphs, eg the plural morpheme in standard English is currently realised by the morph {-s} but archaic and dialectal forms...
  • Body Fluids - pitt.edu

    Body Fluids - pitt.edu

    Functions of AF Physical protection to the fetus Medium for exchange of various chemical Aqueous Humor Fluid that fills the interior chamber of eye Secreted by ciliary body, enters anterior chamber Blockade in the flow of aqueous humor causes glaucoma...
  • PowerPoint Lecture Presentations prepared by Mindy Miller-Kittrell, North

    PowerPoint Lecture Presentations prepared by Mindy Miller-Kittrell, North

    Nucleus Bacterium Bright field Dark field Phase contrast Nomarski Antibodies Bacterium Cell-surface antigens Bacterial cell with bound antibodies carrying dye Fluorescent dye Antibodies carrying dye Light Microscopy PLAY Lamp Light microscope (upside down) Column of transmission electron microscope Lamp Condenser...
  • Unit 10: 2-Dimensional Geometry

    Unit 10: 2-Dimensional Geometry

    Vocabulary . Pyramid - A polyhedron whose base is a polygon and whose lateral faces are triangles that share a common vertex. The pyramid is named by the shape at the base. Square-Based Pyramid. Triangle-Based Pyramid. Hexagonal -Based Pyramid
  • Junior Parent Information Night

    Junior Parent Information Night

    Naviance Student: Overview. Naviance is a web based system designed especially for students and parents to help in making decisions about courses, colleges, and careers. Why is it important to use Naviance Student? Review your child's college and career planning,...
  • This webinar is brought to you by: Why

    This webinar is brought to you by: Why

    Catersource has assembled hundreds of vendors that are not only catering specific but are offering amazing discounts at the show. You will actually save money by shopping at the Catersource tradeshow every year! ... PowerPoint Presentation Last modified by:
  • Payfirma HQ Mobile App User Guide Whats Coming

    Payfirma HQ Mobile App User Guide Whats Coming

    If the app crashes or freezes, close the Payfirma HQ app out. Double tap the home button of your iOS device. Tap and hold the Payfirma HQ icon until a (-) icon appears on the top left corner. Tap on...
  • Bienvenue À MDJH

    Bienvenue À MDJH

    Madame Quann 2015-2016