...

Commits (62)
 Contributors to date (in chronological order, with current affiliations) - Ning Liu, IBM - Jason Cope, DDN - Philip Carns, Argonne National Labs - Misbah Mubarak, Argonne National Labs - Shane Snyder, Argonne National Labs - Jonathan P. Jenkins - Noah Wolfe, RPI - Nikhil Jain, Lawrence Livermore Labs - Matthieu Dorier, Argonne National Labs - Caitlin Ross, RPI - Xu Yang, Amazon Contributors to date (with affiliations at time of contribution) - Philip Carns, Argonne National Laboratory - Misbah Mubarak, Argonne National Laboratory - Shane Snyder, Argonne National Laboratory - Jonathan P. Jenkins, Argonne National Laboratory - Noah Wolfe, Rensselaer Polytechnic Institute - Nikhil Jain, Lawrence Livermore National Laboratory - Jens Domke, Univ. of Dresden - Giorgis Georgakoudis, Lawrence Livermore National Laboratory - Matthieu Dorier, Argonne National Laboratory - Caitlin Ross, Rennselaer Polytechnic Institute - Xu Yang, Illinois Institute of Tech. - Jens Domke, Tokyo Institute of Tech. - Xin Wang, IIT - Xin Wang, Illinois Institute of Tech. - Neil McGlohon, Rensselaer Polytechnic Institute - Elsa Gonsiorowski, Rensselaer Polytechnic Institute - Justin M. Wozniak, Argonne National Laboratory - Robert B. Ross, Argonne National Laboratory - Lee Savoie, Univ. of Arizona - Ning Liu, Rensselaer Polytechnic Institute - Jason Cope, Argonne National Laboratory Contributions: Contributions of external (non-Argonne) collaborators: Misbah Mubarak (ANL) - Introduced 1-D dragonfly and enhanced torus network model. - Added quality of service in dragonfly and megafly network models. - Added MPI simulation layer to simulate MPI operations. - Updated and merged burst buffer storage model with 2-D dragonfly. - Added and validated 2-D dragonfly network model. - Added multiple workload sources including MPI communication, Scalable Workload Models, DUMPI communication traces. - Added online simulation capability with Argobots and SWMs. - Instrumented the network models to report time-stepped series statistics. - Bug fixes for network, storage and workload models with CODES. Neil McGlohon (RPI) - Introduced Dragonfly Plus/Megafly network model. - Merged 1-D dragonfly and 2-D dragonfly network models. - Updated adaptive routing in megafly and 1-D dragonfly network models. - Extended slim fly network model's dual-rail mode to arbitrary number of rails (pending). Nikhil Jain, Abhinav Bhatele (LLNL) - Improvements in credit-based flow control of CODES dragonfly and torus network models. ... ... @@ -29,6 +55,12 @@ Jens Domke (U. of Dresden) - Static routing in fat tree network model including ground work for dumping the topology and reading the routing tables. John Jenkins - Introduced storage models in a separate codes-storage-repo. - Enhanced the codes-mapping APIs to map advanced combinations on PEs. - Bug fixing with network models. - Bug fixing with MPI simulation layer. Xu Yang (IIT) - Added support for running multiple application workloads with CODES MPI Simulation layer, along with supporting scripts and utilities. ... ... @@ -39,6 +71,7 @@ Noah Wolfe (RPI): - Added a fat tree network model that supports full and pruned fat tree network. - Added a multi-rail implementation for the fat tree networks (pending). - Added a dual-rail implementation for slim fly networks (pending). - Bug reporter for CODES network models. Caitlin Ross (RPI): ... ...
COPYRIGHT deleted 100644 → 0
 COPYRIGHT The following is a notice of limited availability of the code, and disclaimer which must be included in the prologue of the code and in all source listings of the code. Copyright Notice + 2013 University of Chicago Permission is hereby granted to use, reproduce, prepare derivative works, and to redistribute to others. This software was authored by: Mathematics and Computer Science Division Argonne National Laboratory, Argonne IL 60439 (and) Computer Science Department Rensselaer Polytechnic Institute, Troy NY 12180 GOVERNMENT LICENSE Portions of this material resulted from work developed under a U.S. Government Contract and are subject to the following license: the Government is granted for itself and others acting on its behalf a paid-up, nonexclusive, irrevocable worldwide license in this computer software to reproduce, prepare derivative works, and perform publicly and display publicly. DISCLAIMER This computer code material was prepared, in part, as an account of work sponsored by an agency of the United States Government. Neither the United States, nor the University of Chicago, nor any of their employees, makes any warranty express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.
LICENSE 0 → 100644
 ************** Copyright © 2019, UChicago Argonne, LLC *************** All Rights Reserved Software Name: CO-Design of Exascale Storage and Network Architectures (CODES) By: Argonne National Laboratory, Rensselaer Polytechnic Institute, Lawrence Livermore National Laboratory, and Illinois Institute of Technology OPEN SOURCE LICENSE Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. ****************************************************************************************************** DISCLAIMER THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ***************************************************************************************************
LICENSE.md 0 → 100644
 ************** Copyright © 2019, UChicago Argonne, LLC *************** All Rights Reserved Software Name: CO-Design of Exascale Storage and Network Architectures (CODES) By: Argonne National Laboratory, Rensselaer Polytechnic Institute, Lawrence Livermore National Laboratory, and Illinois Institute of Technology OPEN SOURCE LICENSE Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 3. Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. ****************************************************************************************************** DISCLAIMER THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ***************************************************************************************************
 # **IMPORTANT NOTICE** ## THE CODES PROJECT HAS BEEN MOVED https://github.com/codes-org/codes As of June 21, 2019, the location of the repo has been moved to GitHub as part of the codebase's shift to its open source BSD License. All issues will be moved in some form to the new location followed by merge requests (if possible). I will do my best to keep the master branch of the two repositories up to date with each other. That being said, if there is any conflict, the GitHub repo should be taken as the truth. Thank you, Neil McGlohon https://github.com/codes-org/codes # CODES Discrete-event Simulation Framework https://xgitlab.cels.anl.gov/codes/codes/wikis/home Discrete event driven simulation of HPC system architectures and subsystems has emerged as a productive and cost-effective means to evaluating potential HPC designs, along with capabilities for executing simulations of extreme scale systems. The goal of the CODES project is use highly parallel simulation to explore the design of exascale storage/network architectures and distributed data-intensive science facilities. Discrete event driven simulation of HPC system architectures and subsystems has emerged as a productive and cost-effective means to evaluating potential HPC designs, along with capabilities for executing simulations of extreme scale systems. The goal of the CODES project is to use highly parallel simulation to explore the design of exascale storage/network architectures and distributed data-intensive science facilities. Our simulations build upon the Rensselaer Optimistic Simulation System (ROSS), a discrete event simulation framework that allows simulations to be run in parallel, decreasing the simulation run time of massive simulations to hours. We are using ROSS to explore topics including large-scale storage systems, I/O workloads, HPC network fabrics, distributed science systems, and data-intensive computation environments. ... ...
 ## README for using ROSS instrumentation with CODES For details about the ROSS instrumentation, see the [ROSS Instrumentation blog post](http://carothersc.github.io/ROSS/instrumentation/instrumentation.html) For details about the ROSS instrumentation, see the [ROSS Instrumentation blog post](http://ross-org.github.io/instrumentation/instrumentation.html) on the ROSS webpage. There are currently 4 types of instrumentation: GVT-based, real time sampling, virtual time sampling, and event tracing. See the ROSS documentation for more info on the specific options or use --help with your model. To collect data about the simulation engine, no changes are needed to model code for any of the instrumentation modes. Some additions to the model code is needed in order to turn on any model-level data collection. See the "Model-level data sampling" section on [ROSS Instrumentation blog post](http://carothersc.github.io/ROSS/instrumentation/instrumentation.html). See the "Model-level data sampling" section on [ROSS Instrumentation blog post](http://ross-org.github.io/instrumentation/instrumentation.html). Here we describe CODES specific details. ### Register Instrumentation Callback Functions ... ... @@ -37,13 +37,13 @@ The second pointer is for the data to be sampled at the GVT or real time samplin In this case the LPs have different function pointers since we want to collect different types of data for the two LP types. For the terminal, I set the appropriate size of the data to be collected, but for the router, the size of the data is dependent on the radix for the dragonfly configuration being used, which isn't known until runtime. *Note*: You can only reuse the function for event tracing for LPs that use the same type of message struct. *Note*: You can only reuse the function for event tracing for LPs that use the same type of message struct. For example, the dragonfly terminal and router LPs both use the terminal_message struct, so they can use the same functions for event tracing. However the model net base LP uses the model_net_wrap_msg struct, so it gets its own event collection function and st_trace_type struct, in order to read the event type correctly from the model. use the same functions for event tracing. However the model net base LP uses the model_net_wrap_msg struct, so it gets its own event collection function and st_trace_type struct, in order to read the event type correctly from the model. In the ROSS instrumentation documentation, there are two methods provided for letting ROSS know about these st_model_types structs. In CODES, this step is a little different, as codes_mapping_setup() calls tw_lp_settype(). In the ROSS instrumentation documentation, there are two methods provided for letting ROSS know about these st_model_types structs. In CODES, this step is a little different, as codes_mapping_setup() calls tw_lp_settype(). Instead, you add a function to return this struct for each of your LP types: C static const st_model_types *dragonfly_get_model_types(void) ... ... @@ -73,7 +73,7 @@ static void router_register_model_types(st_model_types *base_type) At this point, there are two different steps to follow depending on whether the model is one of the model-net models or not. ##### Model-net Models In the model_net_method struct, two fields have been added: mn_model_stat_register and mn_get_model_stat_types. In the model_net_method struct, two fields have been added: mn_model_stat_register and mn_get_model_stat_types. You need to set these to the functions described above. For example: C ... ... @@ -109,27 +109,27 @@ st_model_types svr_model_types[] = { static void svr_register_model_types() { st_model_type_register("server", &svr_model_types[0]); st_model_type_register("ns-lp", &svr_model_types[0]); } int main(int argc, char **argv) { // ... some set up removed for brevity model_net_register(); svr_add_lp_type(); if (g_st_ev_trace || g_st_model_stats) svr_register_model_types(); codes_mapping_setup(); //... }  g_st_ev_trace is a ROSS flag for determining if event tracing is turned on and g_st_model_stats` determines if the GVT-based or real time instrumentation modes are collecting model-level data as well. modes are collecting model-level data as well. ### CODES LPs that currently have event type collection implemented: ... ... @@ -144,4 +144,3 @@ If you're using any of the following CODES models, you don't have to add anythin - slimfly router and terminal LPs (slimfly.c) - fat tree switch and terminal LPs (fat-tree.c) - model-net-base-lp (model-net-lp.c)
 ... ... @@ -2,7 +2,7 @@ # Process this file with autoconf to produce a configure script. AC_PREREQ([2.67]) AC_INIT([codes], [1.0.0], [http://trac.mcs.anl.gov/projects/codes/newticket],[],[http://www.mcs.anl.gov/projects/codes/]) AC_INIT([codes], [1.1.0], [http://trac.mcs.anl.gov/projects/codes/newticket],[],[http://www.mcs.anl.gov/projects/codes/]) LT_INIT ... ...
 ... ... @@ -2,12 +2,12 @@ NOTE: see bottom of this file for suggested configurations on particular ANL machines. 0 - Checkout, build, and install the trunk version of ROSS (https://github.com/carothersc/ROSS). At the time of (https://github.com/ross-org/ROSS). At the time of release (0.6.0), ROSS's latest commit hash was 10d7a06b2d, so this revision is "safe" in the unlikely case incompatible changes come along in the future. If working from the CODES master branches, use the ROSS master branch. git clone http://github.com/carothersc/ROSS.git git clone http://github.com/ross-org/ROSS.git # if using 0.5.2 release: git checkout d3bdc07 cd ROSS mkdir build ... ... @@ -22,7 +22,7 @@ working from the CODES master branches, use the ROSS master branch. ROSS/install/ directory> For more details on installing ROSS, go to https://github.com/carothersc/ROSS/blob/master/README.md . https://github.com/ross-org/ROSS/blob/master/README.md . If using ccmake to configure, don't forget to set CMAKE_C_COMPILER and CMAKE_CXX_COMPILER to mpicc/mpicxx ... ...
 ... ... @@ -14,7 +14,7 @@ https://lists.mcs.anl.gov/mailman/listinfo/codes-ross-users == CODES * main site: http://www.mcs.anl.gov/projects/codes/ * main site: http://www.mcs.anl.gov/projects/codes/ * repositories: * "base" (this repository): git.mcs.anl.gov:radix/codes-base * codes-net (networking component of CODES): git.mcs.anl.gov:radix/codes-net ... ... @@ -22,11 +22,11 @@ https://lists.mcs.anl.gov/mailman/listinfo/codes-ross-users == ROSS * main site, repository, etc.: https://github.com/carothersc/ROSS * main site, repository, etc.: https://github.com/ross-org/ROSS * both the site and repository contain good documentation as well - refer to it for an in-depth introduction and overview of ROSS proper = Components of CODES = Components of CODES == Configuration ... ... @@ -36,7 +36,7 @@ structured configuration file. The configuration format allows categories, and optionally subgroups within the category, of key-value pairs for configuration. The LPGROUPS category defines the LP configuration. The PARAMS category is currently used for networking and ROSS-specific parameters. User-defined categories can also be used. categories can also be used. The configuration system additionally allows LP specialization via the usage of "annotations". This allows two otherwise identical LPs to have different ... ... @@ -125,7 +125,7 @@ The format of the metadata file is a set of lines containing: where: * is the ID of this group (see restrictions) * and form the range of logical client IDs that will * and form the range of logical client IDs that will perform the given workload. Note that the end ID is inclusive, so a start, end pair of 0, 3 will include IDs 0, 1, 2, and 3. An of -1 indicates to use the remaining number of clients as specified by the user. ... ... @@ -202,6 +202,15 @@ should be everything up to the rank number. E.g., if the dumpi files are of the form "dumpi-YYYY.MM.DD.HH.MM.SS-XXXX.bin", then the input should be "dumpi-YYYY.MM.DD.HH.MM.SS-" === Quality of Service Two models (dragonfly-dally.C and dragonfly-plus.C) can now support traffic differentiation and prioritization. The models support quality of service by directing the network traffic on separate class of virtual channels. Additional documentation on using traffic classes can be found at the wiki link: https://xgitlab.cels.anl.gov/codes/codes/wikis/Quality-of-Service === Workload generator helpers The codes-jobmap API (codes/codes-jobmap.h) specifies mechanisms to initialize ... ... @@ -288,7 +297,7 @@ lsm } The API can be found at codes/local-storage-model.h and example usage can be seen in tests/local-storage-model-test.c and tests/conf/lsm-test.conf. seen in tests/local-storage-model-test.c and tests/conf/lsm-test.conf. The queueing policy of LSM is currently FIFO, and the default mode uses an implicit queue, simply incrementing counters and scheduling future events when ... ... @@ -359,7 +368,7 @@ model-net LPs, like any other LP in CODES/ROSS, require specification in the configuration file as well as LP-specific parameters. To identify model-net LPs to CODES, we use the naming scheme of prefixing model-net LP names with the string "modelnet_". For example, the "simplenet" LP (see "model-net models") would be specified in the LPGROUPS section as "modelnet_simplenet". would be specified in the LPGROUPS section as "modelnet_simplenet". Currently, model-net LPs expect their configuration values in the PARAMS section. This is due to historical precedent and may be changed in the future ... ... @@ -389,7 +398,7 @@ program, they will be printed to the specified directory. = model-net models Currently, model-net contains a combination of analytical models and specific models for high-performance networking, with an HPC bent. models for high-performance networking, with an HPC bent. Configuration files for each model can be found in tests/conf, under the name "model-net-test*.conf". ... ... @@ -402,7 +411,7 @@ infinite packet buffering when routing. These are best used for models that require little fidelity out of the network performance. SimpleP2P is the same model as Simplenet, except it provides heterogeneous link capacities from point to point. Rather than a single entry, it requires files containing a matrix of point-to-point bandwidths and latencies. point-to-point bandwidths and latencies. Simplenet models require two configuration parameters: "net_startup_ns" and "net_bw_mbps", which define the startup and bandwidth costs in nanoseconds and ... ... @@ -426,7 +435,7 @@ The only configuration entry the LogGP model requires is configuration file. For more details on gathering parameters for the LogGP model, as well as it's usage and caveats, see the document src/model-net/doc/README.loggp.txt. usage and caveats, see the document src/model-net/doc/README.loggp.txt. == Torus ... ... @@ -454,7 +463,7 @@ The configuration parameters are a little trickier here, as additional LPs other than the "modelnet_dragonfly" LP must be specified. "modelnet_dragonfly" represent the node LPs (terminals), while a second type "dragonfly_router" represents a physical router. At least one "dragonfly_router" LP must be present in every LP group with a "modelnet_dragonfly" LP. present in every LP group with a "modelnet_dragonfly" LP. Further configuration and model setup can be found at src/model-net/doc/README.dragonfly.txt. ... ... @@ -509,7 +518,7 @@ section of the example.conf config file. == Server state and event handlers The server LP state maintains a count of the number of remote messages it has sent and received as well as the number of local completion messages. sent and received as well as the number of local completion messages. For the server event message, we have four message types: KICKOFF, REQ, ACK and LOCAL. With a KICKOFF event, each LP sends a message to itself to begin the ... ... @@ -517,7 +526,7 @@ simulation proper. To avoid event ties, we add a small amount of random noise using codes_local_latency. The REQ message is sent by a server to its neighboring server and when received, neighboring server sends back a message of type ACK. We've shown a hard-coded direct communication method which directly computes the LP ID, and a codes-mapping API-based method. directly computes the LP ID, and a codes-mapping API-based method. == Server reverse computation ... ... @@ -533,19 +542,19 @@ conservative modes, so reverse computation may not be necessary if the simulation is not compute- or memory-intensive. For our example program, recall the "forward" event handlers. They perform the following: following: * Kickoff: send a message to the peer server, and increment sender LP's count of sent messages. * Request (received from peer server): increment receiver count of received messages, and send an acknowledgement to the sender. * Acknowledgement (received from message receiver): send the next message to the receiver and increment messages sent count. Set a flag indicating whether a message has been sent. indicating whether a message has been sent. * Local model-net callback: increment the local model-net received messages count. In terms of LP state, the four operations are simply modifying counts. Hence, the "reverse" event handlers need to merely roll back those changes: the "reverse" event handlers need to merely roll back those changes: * Kickoff: decrement sender LP's count of sent messages. * Request (received from peer server): decrement receiver count of received messages. ... ...
 ... ... @@ -16,7 +16,7 @@ per compute node or one multi-port NIC per node. Adding a generic template for building new network models. For simplest case, only 2 functions and premable changes should suffice to add a new network. Updated Express Mesh network model to serve as an example. For details, see Updated Express Mesh network model to serve as an example. For details, see Darshan workload generator has been updated to use Darshan version 3.x. ... ... @@ -28,11 +28,11 @@ https://xgitlab.cels.anl.gov/codes/codes/wikis/Using-ROSS-Instrumentation-with-C Compatible with ROSS version that enables statistics collection of simulation performance. For details see: http://carothersc.github.io/ROSS/instrumentation/instrumentation.html http://ross-org.github.io/instrumentation/instrumentation.html Online workload replay functionality has been added that allows SWM workloads to be simulated insitu on the network models. WIP to integrate Conceptual domain specific language for network communication. domain specific language for network communication. Multiple traffic patterns were added in the background traffic generation including stencil, all-to-all and random permutation. ... ... @@ -93,7 +93,7 @@ Background network communication using uniform random workload can now be generated. The traffic generation gets automatically shut off when the main workload finishes. Collectives can now be translated into point to point using the CoRTex library. Collectives can now be translated into point to point using the CoRTex library. Performance of MPI_AllReduce is reported when debug_cols option is enabled. ... ...
 ... ... @@ -121,7 +121,7 @@ xleftmargin=6ex % IEEEtran.cls handling of captions and this will result in nonIEEE style % figure/table captions. To prevent this problem, be sure and preload % caption.sty with its "caption=false" package option. This is will preserve % IEEEtran.cls handing of captions. Version 1.3 (2005/06/28) and later % IEEEtran.cls handing of captions. Version 1.3 (2005/06/28) and later % (recommended due to many improvements over 1.2) of subfig.sty supports % the caption=false option directly: %\usepackage[caption=false,font=footnotesize]{subfig} ... ... @@ -188,7 +188,7 @@ easily shared and reused. It also includes a few tips to help avoid common simulation bugs. For more information, ROSS has a bunch of documentation available in their repository/wiki - see \url{https://github.com/carothersc/ROSS}. repository/wiki - see \url{https://github.com/ross-org/ROSS}. \end{abstract} \section{CODES: modularizing models} ... ... @@ -394,7 +394,7 @@ action upon the completion of them. More generally, the problem is: an event issuance (an ack to the client) is based on the completion of more than one asynchronous/parallel events (local write on primary server, forwarding write to replica server). Further complicating the matter for storage simulations, there can be any number of outstanding requests, each waiting on multiple events. can be any number of outstanding requests, each waiting on multiple events. In ROSS's sequential and conservative parallel modes, the necessary state can easily be stored in the LP as a queue of statuses for each set of events, ... ... @@ -488,7 +488,7 @@ Most core ROSS examples are design to intentionally hit the end timestamp for the simulation (i.e. they are modeling a continuous, steady state system). This isn't necessarily true for other models. Quite simply, set g\_tw\_ts\_end to an arbitrary large number when running simulations that have a well-defined end-point in terms of events processed. that have a well-defined end-point in terms of events processed. Within the LP finalize function, do not call tw\_now. The time returned may not be consistent in the case of an optimistic simulation. ... ... @@ -515,7 +515,7 @@ section(s). \item generating multiple concurrent events makes rollback more difficult \end{enumerate} \item use dummy events to work around "event-less" advancement of simulation time \item use dummy events to work around "event-less" advancement of simulation time \item add a small amount of time "noise" to events to prevent ties ... ...
 ... ... @@ -44,7 +44,7 @@ Notes on how to release a new version of CODES 4. Upload the release tarball - Our release directory is at ftp.mcs.anl.gov/pub/CODES/releases . There's no web interface, so you have to get onto an MCS workstation and copy the release in that way (the ftp server is mounted at /homes/ftp). release in that way (the ftp server is mounted at /mcs/ftp.mcs.anl.gov). 5. Update website - Project wordpress: http://www.mcs.anl.gov/projects/codes/ (you need ... ...
 ... ... @@ -246,7 +246,7 @@ int main( /* calculate the number of servers in this simulation, * ignoring annotations */ num_servers = codes_mapping_get_lp_count(group_name, 0, "server", NULL, 1); num_servers = codes_mapping_get_lp_count(group_name, 0, "nw-lp", NULL, 1); /* for this example, we read from a separate configuration group for * server message parameters. Since they are constant for all LPs, ... ... @@ -273,7 +273,7 @@ static void svr_add_lp_type() { /* lp_type_register should be called exactly once per process per * LP type */ lp_type_register("server", svr_get_lp_type()); lp_type_register("nw-lp", svr_get_lp_type()); } static void svr_init( ... ...
 ... ... @@ -3,14 +3,14 @@ # of application- and codes-specific key-value pairs. LPGROUPS { # in our simulation, we simply have a set of servers, each with # in our simulation, we simply have a set of servers (nw-lp), each with # point-to-point access to each other SERVERS { # required: number of times to repeat the following key-value pairs repetitions="16"; # application-specific: parsed in main server="1"; nw-lp="1"; # model-net-specific field defining the network backend. In this example, # each server has one NIC, and each server are point-to-point connected modelnet_simplenet="1"; ... ...
 ... ... @@ -18,6 +18,7 @@ argobots_libs=@ARGOBOTS_LIBS@ argobots_cflags=@ARGOBOTS_CFLAGS@ swm_libs=@SWM_LIBS@ swm_cflags=@SWM_CFLAGS@ swm_datarootdir=@SWM_DATAROOTDIR@ Name: codes-base Description: Base functionality for CODES storage simulation ... ... @@ -25,4 +26,4 @@ Version: @PACKAGE_VERSION@ URL: http://trac.mcs.anl.gov/projects/CODES Requires: Libs: -L${libdir} -lcodes${ross_libs} ${argobots_libs}${swm_libs} ${darshan_libs}${dumpi_libs} ${cortex_libs} Cflags: -I${includedir} ${swm_datarootdir}${ross_cflags} ${darshan_cflags}${swm_cflags} ${argobots_cflags}${dumpi_cflags} ${cortex_cflags} Cflags: -I${includedir} -I${swm_datarootdir}${ross_cflags} ${darshan_cflags}${swm_cflags} ${argobots_cflags}${dumpi_cflags} \${cortex_cflags}
This diff is collapsed.
 ... ... @@ -4,7 +4,7 @@ # In hindsight this was a lot more complicated than I intended. It was looking to solve a complex problem that turned out to be invalid from the beginning. ### USAGE ### # Correct usage: python3 script.py # Correct usage: python3 dragonfly-plus-topo-gen-v2.py ### ### import sys ... ... @@ -573,37 +573,37 @@ def mainV3(): print(A.astype(int)) def mainV2(): if(len(argv) < 8): raise Exception("Correct usage: python %s " % sys.argv[0]) # def mainV2(): # if(len(argv) < 8): # raise Exception("Correct usage: python %s " % sys.argv[0]) num_groups = int(argv[1]) num_spine_pg = int(argv[2]) num_leaf_pg = int(argv[3]) router_radix = int(argv[4]) term_per_leaf = int(argv[5]) intra_filename = argv[6] inter_filename = argv[7] # num_groups = int(argv[1]) # num_spine_pg = int(argv[2]) # num_leaf_pg = int(argv[3]) # router_radix = int(argv[4]) # term_per_leaf = int(argv[5]) # intra_filename = argv[6] # inter_filename = argv[7] parseOptionArguments() # parseOptionArguments() dfp_network = DragonflyPlusNetwork(num_groups, num_spine_pg, num_leaf_pg, router_radix, num_hosts_per_leaf=term_per_leaf) # dfp_network = DragonflyPlusNetwork(num_groups, num_spine_pg, num_leaf_pg, router_radix, num_hosts_per_leaf=term_per_leaf) if not DRYRUN: dfp_network.writeIntraconnectionFile(intra_filename) dfp_network.writeInterconnectionFile(inter_filename) # if not DRYRUN: # dfp_network.writeIntraconnectionFile(intra_filename) # dfp_network.writeInterconnectionFile(inter_filename) if LOUDNESS is not Loudness.QUIET: print("\nNOTE: THIS STILL CAN'T DO THE MED-LARGE TOPOLOGY RIGHT\n") # if LOUDNESS is not Loudness.QUIET: # print("\nNOTE: THIS STILL CAN'T DO THE MED-LARGE TOPOLOGY RIGHT\n") print(dfp_network.getSummary()) # print(dfp_network.getSummary()) if SHOW_ADJACENCY == 1: print("\nPrinting Adjacency Matrix:") # if SHOW_ADJACENCY == 1: # print("\nPrinting Adjacency Matrix:") np.set_printoptions(linewidth=400,threshold=10000,edgeitems=200) A = dfp_network.getAdjacencyMatrix(AdjacencyType.ALL_CONNS) print(A.astype(int)) # np.set_printoptions(linewidth=400,threshold=10000,edgeitems=200) # A = dfp_network.getAdjacencyMatrix(AdjacencyType.ALL_CONNS) # print(A.astype(int)) if __name__ == '__main__': mainV3()
 ... ... @@ -3,14 +3,14 @@ # of application- and codes-specific key-value pairs. LPGROUPS { # in our simulation, we simply have a set of servers, each with # in our simulation, we simply have a set of servers (nw-lp), each with # point-to-point access to each other SERVERS { # required: number of times to repeat the following key-value pairs repetitions="C_NUM_SERVERS"; # application-specific: parsed in main server="1"; nw-lp="1"; # model-net-specific field defining the network backend. In this example, # each server has one NIC, and each server are point-to-point connected modelnet_simplenet="1"; ... ...
 ... ... @@ -41,7 +41,7 @@ PARAMS # bandwidth in GiB/s for compute node-router channels cn_bandwidth="16.0"; # ROSS message size message_size="656"; message_size="736"; # number of compute nodes connected to router, dictated by dragonfly config # file num_cns_per_router="2"; ... ...
 ... ... @@ -3,7 +3,7 @@ LPGROUPS MODELNET_GRP { repetitions="198"; server="384"; nw-lp="384"; modelnet_fattree="24"; fattree_switch="6"; } ... ...
 ... ... @@ -3,7 +3,7 @@ LPGROUPS MODELNET_GRP { repetitions="252"; server="288"; nw-lp="288"; modelnet_fattree="18"; fattree_switch="6"; } ... ...
 ... ... @@ -23,6 +23,6 @@ PARAMS local_bandwidth="5.25"; global_bandwidth="4.7"; cn_bandwidth="5.25"; message_size="656"; message_size="736"; routing="adaptive"; }
 ... ... @@ -12,7 +12,7 @@ PARAMS { ft_type="0"; packet_size="512"; message_size="592"; message_size="736"; chunk_size="512"; modelnet_scheduler="fcfs"; #modelnet_scheduler="round-robin"; ... ...
 ... ... @@ -31,6 +31,6 @@ PARAMS cn_bandwidth="9.0"; router_delay="0"; link_delay="0"; message_size="656"; message_size="736"; routing="minimal"; }
 ... ... @@ -10,7 +10,7 @@ LPGROUPS PARAMS { packet_size="512"; message_size="656"; message_size="736"; modelnet_order=( "torus" ); # scheduler options modelnet_scheduler="fcfs"; ... ...
 ... ... @@ -3,7 +3,7 @@ LPGROUPS MODELNET_GRP { repetitions="264"; server="4"; nw-lp="4"; modelnet_dragonfly="4"; modelnet_dragonfly_router="1"; } ... ...
 ... ... @@ -3,7 +3,7 @@ LPGROUPS MODELNET_GRP { repetitions="198"; # repetitions = Ne = total # of edge switches. For type0 Ne = Np*Ns = ceil(N/Ns*(k/2))*(k/2) = ceil(N/(k/2)^2)*(k/2) server="18"; nw-lp="18"; modelnet_fattree="18"; fattree_switch="3"; } ... ...
 ... ... @@ -3,7 +3,7 @@ LPGROUPS MODELNET_GRP { repetitions="32"; # repetitions = Ne = total # of edge switches. For type0 Ne = Np*Ns = ceil(N/Ns*(k/2))*(k/2) = ceil(N/(k/2)^2)*(k/2) server="4"; nw-lp="4"; modelnet_fattree="4"; fattree_switch="3"; } ... ...
 ... ... @@ -3,7 +3,7 @@ LPGROUPS MODELNET_GRP { repetitions="50"; server="3"; nw-lp="3"; modelnet_slimfly="3"; slimfly_router="1"; } ... ...
This diff is collapsed.
 ... ... @@ -165,7 +165,7 @@ static const st_model_types *ft_svr_get_model_stat_types(void) void ft_svr_register_model_stats() { st_model_type_register("server", ft_svr_get_model_stat_types()); st_model_type_register("nw-lp", ft_svr_get_model_stat_types()); } const tw_optdef app_opt [] = ... ... @@ -184,7 +184,7 @@ const tw_lptype* svr_get_lp_type() static void svr_add_lp_type() { lp_type_register("server", svr_get_lp_type()); lp_type_register("nw-lp", svr_get_lp_type()); } static void issue_event( ... ... @@ -477,7 +477,7 @@ int main( MPI_Finalize(); return 0; } num_servers_per_rep = codes_mapping_get_lp_count("MODELNET_GRP", 1, "server", num_servers_per_rep = codes_mapping_get_lp_count("MODELNET_GRP", 1, "nw-lp", NULL, 1); configuration_get_value_int(&config, "PARAMS", "num_routers", NULL, &num_routers_per_grp); ... ... @@ -485,7 +485,7 @@ int main( num_nodes = num_groups * num_routers_per_grp * (num_routers_per_grp / 2); num_nodes_per_grp = num_routers_per_grp * (num_routers_per_grp / 2); num_nodes = codes_mapping_get_lp_count("MODELNET_GRP", 0, "server", NULL, 1); num_nodes = codes_mapping_get_lp_count("MODELNET_GRP", 0, "nw-lp", NULL, 1); printf("num_nodes:%d \n",num_nodes); ... ...
 ... ... @@ -160,7 +160,7 @@ static const st_model_types *svr_get_model_stat_types(void) void svr_register_model_types() { st_model_type_register("server", svr_get_model_stat_types()); st_model_type_register("nw-lp", svr_get_model_stat_types()); } const tw_optdef app_opt [] = ... ... @@ -181,7 +181,7 @@ const tw_lptype* svr_get_lp_type() static void svr_add_lp_type() { lp_type_register("server", svr_get_lp_type()); lp_type_register("nw-lp", svr_get_lp_type()); } /* convert GiB/s and bytes to ns */ ... ... @@ -523,7 +523,7 @@ int main( net_id = *net_ids; free(net_ids); num_servers_per_rep = codes_mapping_get_lp_count("MODELNET_GRP", 1, "server", NULL, 1); num_servers_per_rep = codes_mapping_get_lp_count("MODELNET_GRP", 1, "nw-lp", NULL, 1); configuration_get_value_int(&config, "PARAMS", "num_terminals", NULL, &num_terminals); configuration_get_value_int(&config, "PARAMS", "num_routers", NULL, &num_routers_per_grp); num_groups = (num_routers_per_grp * 2); ... ...
 ... ... @@ -147,7 +147,7 @@ static const st_model_types *svr_get_model_stat_types(void) void svr_register_model_types() { st_model_type_register("server", svr_get_model_stat_types()); st_model_type_register("nw-lp", svr_get_model_stat_types()); } const tw_optdef app_opt [] = ... ... @@ -170,7 +170,7 @@ const tw_lptype* svr_get_lp_type() static void svr_add_lp_type() { lp_type_register("server", svr_get_lp_type()); lp_type_register("nw-lp", svr_get_lp_type()); } static void issue_event( ... ... @@ -448,7 +448,7 @@ int main( MPI_Finalize(); return 0; } num_servers_per_rep = codes_mapping_get_lp_count("MODELNET_GRP", 1, "server", num_servers_per_rep = codes_mapping_get_lp_count("MODELNET_GRP", 1, "nw-lp", NULL, 1); configuration_get_value_int(&config, "PARAMS", "num_routers", NULL, &num_routers_per_grp); ... ...
 ... ... @@ -58,7 +58,7 @@ p*a*g=8*4*33). This configuration can be specified in the config file in the fol MODELNET_GRP { repetitions="264"; server="4"; nw-lp="4"; modelnet_dragonfly="4"; dragonfly_router="1"; } ... ... @@ -70,7 +70,7 @@ PARAMS } The first section, MODELNET_GRP specified the number of LPs and the layout of LPs. In the above case, there are 264 repetitions of 4 server LPs, 4 dragonfly LPs. In the above case, there are 264 repetitions of 4 server LPs (nw-lp), 4 dragonfly network node LPs and 1 dragonfly router LP, which makes a total of 264 routers, 1056 network nodes and 1056 servers in the network. The second section, PARAMS uses 'num_routers' for the dragonfly topology lay out and setsup the ... ...
 ... ... @@ -8,7 +8,7 @@ bare-minimum example config file: MODELNET_GRP { repetitions="12"; server="4"; nw-lp="4"; modelnet_fattree="4"; fattree_switch="3"; } ... ... @@ -24,7 +24,7 @@ PARAMS The first section, MODELNET_GRP specifies the LP types, number of LPs per type and their configuration. In the above case, there are 12 repetitions each with 4 server LPs, 4 fat tree network node/terminal LPs and 3 fat tree switch LPs. Each server LPs (nw-lp), 4 fat tree network node/terminal LPs and 3 fat tree switch LPs. Each repetition represents a leaf level switch, nodes connected to it, and higher level switches that may be needed to construct the fat-tree. The 'fattree_switch' parameter indicates there are 3 levels to this fat tree and ... ...
 ... ... @@ -73,7 +73,7 @@ generator sets. This configuration can be specified in the config file in the fo MODELNET_GRP { repetitions="50"; server="3"; nw-lp="3"; modelnet_slimfly="3"; slimfly_router="1"; } ... ... @@ -88,7 +88,7 @@ PARAMS } The first section, MODELNET_GRP specified the number of LPs and the layout of LPs. In the above case, there are 50 repetitions of 3 server LPs, 3 slim fly LPs. In the above case, there are 50 repetitions of 3 server LPs (nw-lp), 3 slim fly network node LPs and 1 slim fly router LP, which makes a total of 50 routers, 150 network nodes and 150 servers in the network. The second section, PARAMS uses 'num_routers' for the slim fly topology lay out and sets up the ... ...
 ... ... @@ -415,6 +415,8 @@ struct router_state tw_stime* busy_time; tw_stime* busy_time_sample; unsigned long* stalled_chunks; //Counter for when a packet is put into queued messages instead of routing terminal_dally_message_list ***pending_msgs; terminal_dally_message_list ***pending_msgs_tail; terminal_dally_message_list ***queued_msgs; ... ... @@ -1538,7 +1540,9 @@ void router_dally_setup(router_state * r, tw_lp * lp) r->link_traffic_sample = (int64_t*)calloc(p->radix, sizeof(int64_t)); r->cur_hist_num = (int*)calloc(p->radix, sizeof(int)); r->prev_hist_num = (int*)calloc(p->radix, sizeof(int)); r->stalled_chunks = (unsigned long*)calloc(p->radix, sizeof(unsigned long)); r->last_sent_chan = (int*) calloc(p->num_router_rows, sizeof(int)); r->vc_occupancy = (int**)calloc(p->radix , sizeof(int*)); r->in_send_loop = (int*)calloc(p->radix, sizeof(int)); ... ... @@ -1781,9 +1785,14 @@ static void packet_generate_rc(terminal_state * s, tw_bf * bf, terminal_dally_me if(bf->c5) { s->in_send_loop = 0; } if(bf->c11) { if (bf->c11) { s->issueIdle = 0; } if(bf->c8) { s->last_buf_full = msg->saved_busy_time; } } struct mn_stats* stat; stat = model_net_find_stats(msg->category, s->dragonfly_stats_array); stat->send_count--; ... ... @@ -1915,6 +1924,15 @@ static void packet_generate(terminal_state * s, tw_bf * bf, terminal_dally_messa } else { bf->c11 = 1; s->issueIdle = 1; //this block was missing from when QOS was added - readded 5-21-19 if(s->last_buf_full == 0.0) { bf->c8 = 1; msg->saved_busy_time = s->last_buf_full; /* TODO: Assumes a single vc from terminal to router */ s->last_buf_full = tw_now(lp); } } if(s->in_send_loop == 0) { ... ... @@ -2179,8 +2197,7 @@ static void packet_send(terminal_state * s, tw_bf * bf, terminal_dally_message * msg->num_rngs = 0; msg->num_cll = 0; if(num_qos_levels > 1) vcg = get_next_vcg(s, bf, msg, lp); vcg = get_next_vcg(s, bf, msg, lp); /* For a terminal to router connection, there would be as many VCGs as number * of VCs*/ ... ... @@ -2312,6 +2329,7 @@ static void packet_send(terminal_state * s, tw_bf * bf, terminal_dally_message * s->busy_time += (tw_now(lp) - s->last_buf_full); s->busy_time_sample += (tw_now(lp) - s->last_buf_full); s->ross_sample.busy_time_sample += (tw_now(lp) - s->last_buf_full); msg->saved_busy_time_ross = s->busy_time_ross_sample; s->busy_time_ross_sample += (tw_now(lp) - s->last_buf_full); s->last_buf_full = 0.0; } ... ... @@ -3136,11 +3154,11 @@ dragonfly_dally_terminal_final( terminal_state * s, if(s->terminal_id == 0) { written += sprintf(s->output_buf + written, "# Format < dest_type> "); written += sprintf(s->output_buf + written, "# Format < dest_type> "); // fprintf(fp, "# Format <# Flits/Packets finished> \n"); } written += sprintf(s->output_buf + written, "\n%u %s %llu %s %s %llu %lf", s->terminal_id, "T", s->router_id, "R", "CN", LLU(s->total_msg_size), s->busy_time); written += sprintf(s->output_buf + written, "\n%u %s %llu %s %s %llu %lf %d", s->terminal_id, "T", s->router_id, "R", "CN", LLU(s->total_msg_size), s->busy_time, -1); //note that terminals don't have stalled chuncks because of model net scheduling only gives a terminal what it can handle (-1 to show N/A) lp_io_write(lp->gid, (char*)"dragonfly-link-stats", written, s->output_buf); ... ... @@ -3211,36 +3229,40 @@ void dragonfly_dally_router_final(router_state * s, if(d != src_rel_id) { int dest_ab_id = local_grp_id * p->num_routers + d; written += sprintf(s->output_buf + written, "\n%d %s %d %s %s %llu %lf", written += sprintf(s->output_buf + written, "\n%d %s %d %s %s %llu %lf %lu", s->router_id, "R", dest_ab_id, "R", "L", s->link_traffic[d], s->busy_time[d]); s->busy_time[d], s->stalled_chunks[d]); } } map< int, vector > &curMap = interGroupLinks[s->router_id]; map< int, vector >::iterator it = curMap.begin(); for(; it != curMap.end(); it++) { /* TODO: Works only for single global connections right now. Make it functional * for a 2-D dragonfly. */ for(int l = 0; l < it->second.size(); l++) { int dest_rtr_id = it->second[l].dest; int offset = it->second[l].offset; assert(offset >= 0 && offset < p->num_global_channels); written += sprintf(s->output_buf + written, "\n%d %s %d %s %s %llu %lf", s->router_id, "R", dest_rtr_id, "R", "G", s->link_traffic[offset], s->busy_time[offset]); } map< int, vector > &curMap = interGroupLinks[s->router_id]; map< int, vector >::iterator it = curMap.begin(); for(; it != curMap.end(); it++) { /* TODO: Works only for single global connections right now. Make it functional * for a 2-D dragonfly. */ for(int l = 0; l < it->second.size(); l++) { int dest_rtr_id = it->second[l].dest; int offset = it->second[l].offset; assert(offset >= 0 && offset < p->num_global_channels); written += sprintf(s->output_buf + written, "\n%d %s %d %s %s %llu %lf %lu", s->router_id, "R", dest_rtr_id, "R", "G", s->link_traffic[offset], s->busy_time[offset], s->stalled_chunks[offset]); } } sprintf(s->output_buf + written, "\n"); lp_io_write(lp->gid, (char*)"dragonfly-link-stats", written, s->output_buf); ... ... @@ -4012,6 +4034,11 @@ static void router_packet_receive_rc(router_state * s, } } if(bf->c4) { s->stalled_chunks[output_port]--; if(bf->c22) { s->last_buf_full[output_port] = msg->saved_busy_time; } delete_terminal_dally_message_list(return_tail(s->queued_msgs[output_port], s->queued_msgs_tail[output_port], output_chan)); s->queued_count[output_port] -= s->params->chunk_size; ... ... @@ -4282,12 +4309,27 @@ if(cur_chunk->msg.path_type == NON_MINIMAL) } else { bf->c4 = 1; s->stalled_chunks[output_port]++; cur_chunk->msg.saved_vc = msg->vc_index; cur_chunk->msg.saved_channel = msg->output_chan; assert(output_chan < s->params->num_vcs && output_port < s->params->radix); append_to_terminal_dally_message_list( s->queued_msgs[output_port], s->queued_msgs_tail[output_port], output_chan, cur_chunk); s->queued_count[output_port] += s->params->chunk_size; //THIS WAS REMOVED WHEN QOS WAS INSTITUTED - READDED 5/20/19 /* a check for pending msgs is non-empty then we dont set anything. If * that is empty then we check if last_buf_full is set or not. If already * set then we don't overwrite it. If two packets arrive next to each other * then the first person should be setting it. */ if(s->pending_msgs[output_port][output_chan] == NULL && s->last_buf_full[output_port] == 0.0) { bf->c22 = 1; msg->saved_busy_time = s->last_buf_full[output_port]; s->last_buf_full[output_port] = tw_now(lp); } } msg->saved_vc = output_port; ... ... @@ -4395,7 +4437,7 @@ router_packet_send( router_state * s, msg->num_rngs = 0; int num_qos_levels = s->params->num_qos_levels; int output_chan = get_next_router_vcg(s, bf, msg, lp); int output_chan = get_next_router_vcg(s, bf, msg, lp); //includes default output_chan setting functionality if qos not enabled msg->saved_vc = output_port; msg->saved_channel = output_chan; ... ... @@ -4403,7 +4445,7 @@ router_packet_send( router_state * s, if(output_chan < 0) { bf->c1 = 1; s->in_send_loop[output_port] = 0; if(s->queued_count[output_port] && !s->last_buf_full[output_port]) if(s->queued_count[output_port] && !s->last_buf_full[output_port]) //5-21-19, not sure why this was added here with the qos stuff { bf->c2 = 1; msg->saved_busy_time = s->last_buf_full[output_port]; ... ... @@ -4416,7 +4458,7 @@ router_packet_send( router_state * s, assert(cur_entry != NULL); if(s->last_buf_full[output_port]) if(s->last_buf_full[output_port]) //5-12-19, same here as above comment { bf->c8 = 1; msg->saved_rcv_time = s->busy_time[output_port]; ... ...
 ... ... @@ -1643,6 +1643,7 @@ void issue_rtr_bw_monitor_event(router_state *s, tw_bf *bf, terminal_plus_messag s->qos_data[i][j] = 0; } s->busy_time_sample[i] = 0; s->ross_rsample.busy_time[i] = 0; } if(tw_now(lp) > max_qos_monitor) ... ... @@ -2297,6 +2298,7 @@ static void packet_send(terminal_state *s, tw_bf *bf, terminal_plus_message *msg s->busy_time += (tw_now(lp) - s->last_buf_full); s->busy_time_sample += (tw_now(lp) - s->last_buf_full); s->ross_sample.busy_time_sample += (tw_now(lp) - s->last_buf_full); s->last_buf_full = 0.0; } } ... ... @@ -2737,7 +2739,7 @@ void dragonfly_plus_rsample_fin(router_state *s, tw_lp *lp) if (s->router_id == 0) { /* write metadata file */ char meta_fname[64]; sprintf(meta_fname, "dragonfly-plus-router-sampling.meta"); sprintf(meta_fname, "dragonfly-router-sampling.meta"); FILE *fp = fopen(meta_fname, "w"); fprintf(fp, ... ... @@ -2750,7 +2752,7 @@ void dragonfly_plus_rsample_fin(router_state *s, tw_lp *lp) } char rt_fn[MAX_NAME_LENGTH]; if (strcmp(router_sample_file, "") == 0) sprintf(rt_fn, "dragonfly-plus-router-sampling-%ld.bin", g_tw_mynode); sprintf(rt_fn, "dragonfly-router-sampling-%ld.bin", g_tw_mynode); else sprintf(rt_fn, "%s-%ld.bin", router_sample_file, g_tw_mynode); ... ... @@ -4168,6 +4170,7 @@ static void router_packet_send_rc(router_state *s, tw_bf *bf, terminal_plus_mess { s->busy_time[output_port] = msg->saved_rcv_time; s->busy_time_sample[output_port] = msg->saved_sample_time; s->ross_rsample.busy_time[output_port] = msg->saved_sample_time; s->last_buf_full[output_port] = msg->saved_busy_time; } ... ... @@ -4256,6 +4259,7 @@ static void router_packet_send(router_state *s, tw_bf *bf, terminal_plus_message msg->saved_sample_time = s->busy_time_sample[output_port]; s->busy_time[output_port] += (tw_now(lp) - s->last_buf_full[output_port]); s->busy_time_sample[output_port] += (tw_now(lp) - s->last_buf_full[output_port]); s->ross_rsample.busy_time[output_port] += (tw_now(lp) - s->last_buf_full[output_port]); s->last_buf_full[output_port] = 0.0; } ... ... @@ -4330,6 +4334,7 @@ static void router_packet_send(router_state *s, tw_bf *bf, terminal_plus_message bf->c11 = 1; s->link_traffic[output_port] += (cur_entry->msg.packet_size % s->params->chunk_size); s->link_traffic_sample[output_port] += (cur_entry->msg.packet_size % s->params->chunk_size); s->ross_rsample.link_traffic_sample[output_port] += (cur_entry->msg.packet_size % s->params->chunk_size); msg_size = cur_entry->msg.packet_size % s->params->chunk_size; } else { ... ... @@ -4399,6 +4404,7 @@ static void router_buf_update_rc(router_state *s, tw_bf *bf, terminal_plus_messa if (bf->c3) { s->busy_time[indx] = msg->saved_rcv_time; s->busy_time_sample[indx] = msg->saved_sample_time; s->ross_rsample.busy_time[indx] = msg->saved_sample_time; s->last_buf_full[indx] = msg->saved_busy_time; } if (bf->c1) { ... ... @@ -4430,6 +4436,7 @@ static void router_buf_update(router_state *s, tw_bf *bf, terminal_plus_message msg->saved_sample_time = s->busy_time_sample[indx]; s->busy_time[indx] += (tw_now(lp) - s->last_buf_full[indx]); s->busy_time_sample[indx] += (tw_now(lp) - s->last_buf_full[indx]); s->ross_rsample.busy_time[indx] += (tw_now(lp) - s->last_buf_full[indx]); s->last_buf_full[indx] = 0.0; } if (s->queued_msgs[indx][output_chan] != NULL) { ... ...
 ... ... @@ -12,13 +12,13 @@ LPGROUPS MODELNET_GRP { repetitions="16"; server="1"; nw-lp="1"; example_net="1"; } } ------------------------------------------------------------------------------ In this config file, there are multiple LP types defined in a single LP group (As we will see in a later example, there can be multiple LP groups in a config file too). There is 1 server LP and 1 example_net example, there can be multiple LP groups in a config file too). There is 1 server LP (nw-lp) and 1 example_net LP type in a group and this combination is repeated 16 time (repetitions="16"). ROSS will assign the LPs to the PEs (PEs is an abstraction for MPI rank in ROSS) by placing 1 server LP then 1 example_net LP a total of 16 times. This configuration is useful if there is some form of communication involved ... ... @@ -33,7 +33,7 @@ LPGROUPS MODELNET_GRP { repetitions="16"; server="1"; nw-lp="1"; example_net="2"; } } ... ... @@ -51,7 +51,7 @@ Following is an example function that registers 'server' lp type. static void svr_add_lp_type() { lp_type_register("server", svr_get_lp_type()); lp_type_register("nw-lp", svr_get_lp_type()); } Step 4: Call to codes_mapping_setup that sets up the LPs of each registered LP type as specified in the config ... ... @@ -67,7 +67,7 @@ The second argument indicates whether to include the number of repetitions into the count or not (the NULL, 0 arguments will be discussed later). For example, to query the number of server LPs in example-test2.conf file, calling num_servers = codes_mapping_get_lp_count("MODELNET_GRP", 1, "server", NULL, 0); num_servers = codes_mapping_get_lp_count("MODELNET_GRP", 1, "nw-lp", NULL, 0); will return 2. ... ...
 ... ... @@ -462,7 +462,7 @@ static void codes_mapping_init(void) /* have 16 kps per pe, this is the optimized configuration for ROSS custom mapping */ for(kpid = 0; kpid < nkp_per_pe; kpid++) tw_kp_onpe(kpid, g_tw_pe[0]); tw_kp_onpe(kpid, g_tw_pe); tw_lpid lp_start = g_tw_mynode * lps_per_pe_floor + mini(g_tw_mynode,lps_leftover); ... ... @@ -474,7 +474,7 @@ static void codes_mapping_init(void) ross_gid = lpid; ross_lid = lpid - lp_start; kpid = ross_lid % g_tw_nkp; pe = tw_getpe(kpid % g_tw_npe); pe = g_tw_pe; codes_mapping_get_lp_info(ross_gid, NULL, &grp_id, lp_type_name, &lpt_id, NULL, &rep_id, &offset); #if CODES_MAPPING_DEBUG ... ...
 ... ... @@ -375,7 +375,7 @@ static int write_id(char* directory, char* identifier, MPI_Comm comm) i--; buf = buf->next; } MPI_Type_hindexed(id->buffers_count, lengths, displacements, MPI_Type_create_hindexed(id->buffers_count, lengths, displacements, MPI_BYTE, &mtype); MPI_Type_commit(&mtype); free(lengths); ... ...
 ... ... @@ -23,10 +23,11 @@ #include "lammps.h" #include "nekbone_swm_user_code.h" #include "nearest_neighbor_swm_user_code.h" // #include "all_to_one_swm_user_code.h" #include "all_to_one_swm_user_code.h" #define ALLREDUCE_SHORT_MSG_SIZE 2048 //#define DBG_COMM 0 using namespace std; ... ... @@ -36,7 +37,7 @@ static int rank_tbl_pop = 0;