Scalable Hierarchical Aggregation Protocol (SHArP) Release Notes Rev 1.0 www.mellanox.com Mellanox Technologies NOTE: THIS HARDWARE, SOFTWARE OR TEST SUITE PRODUCT (“PRODUCT(S)”) AND ITS RELATED DOCUMENTATION ARE PROVIDED BY MELLANOX TECHNOLOGIES “AS-IS” WITH ALL FAULTS OF ANY KIND AND SOLELY FOR THE PURPOSE OF AIDING THE CUSTOMER IN TESTING APPLICATIONS THAT USE THE PRODUCTS IN DESIGNATED SOLUTIONS. THE CUSTOMER'S MANUFACTURING TEST ENVIRONMENT HAS NOT MET THE STANDARDS SET BY MELLANOX TECHNOLOGIES TO FULLY QUALIFY THE PRODUCT(S) AND/OR THE SYSTEM USING IT. THEREFORE, MELLANOX TECHNOLOGIES CANNOT AND DOES NOT GUARANTEE OR WARRANT THAT THE PRODUCTS WILL OPERATE WITH THE HIGHEST QUALITY. ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT ARE DISCLAIMED. IN NO EVENT SHALL MELLANOX BE LIABLE TO CUSTOMER OR ANY THIRD PARTIES FOR ANY DIRECT, INDIRECT, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES OF ANY KIND (INCLUDING, BUT NOT LIMITED TO, PAYMENT FOR 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 FROM THE USE OF THE PRODUCT(S) AND RELATED DOCUMENTATION EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. Mellanox Technologies 350 Oakmead Parkway Suite 100 Sunnyvale, CA 94085 U.S.A. www.mellanox.com Tel: (408) 970-3400 Fax: (408) 970-3403 © Copyright 2016. Mellanox Technologies LTD. All Rights Reserved. Mellanox®, Mellanox logo, BridgeX®, CloudX logo, CompustorX®, Connect-IB®, ConnectX®, CoolBox®, COREDirect®, EZchip®, EZchip logo, EZappliance®, EZdesign®, EZdriver®, EZsystem®, GPUDirect®, InfiniHost®, InfiniScale®, Kotura®, Kotura logo, Mellanox Federal Systems®, Mellanox Open Ethernet®, Mellanox ScalableHPC®, Mellanox TuneX®, Mellanox Connect Accelerate Outperform logo, Mellanox Virtual Modular Switch®, MetroDX®, MetroX®, MLNX-OS®, NP-1c®, NP-2®, NP-3®, Open Ethernet logo, PhyX®, SwitchX®, Tilera®, Tilera logo, TestX®, TuneX®, The Generation of Open Ethernet logo, UFM®, Virtual Protocol Interconnect®, Voltaire® and Voltaire logo are registered trademarks of Mellanox Technologies, Ltd. All other trademarks are property of their respective owners. For the most updated list of Mellanox trademarks, visit http://www.mellanox.com/page/trademarks 2 Mellanox Technologies Table of Contents Rev 1.0 Table of Contents Document Revision History.................................................................................................................. 5 1 Overview .......................................................................................................................................... 6 1.1 Supported OS and Platforms .................................................................................................. 6 2 Main Features in This Release ....................................................................................................... 7 3 Limitations ....................................................................................................................................... 8 3 Mellanox Technologies Rev 1.0 Table of Contents List of Tables Table 1: Document Revision History ....................................................................................................... 5 Table 2: New Features, Changes and Fixes ........................................................................................... 7 4 Mellanox Technologies Scalable Hierarchical Aggregation Protocol (SHArP) Release Notes Document Revision History Table 1: Document Revision History Revision Date Description 1.0 June 2016 Initial version of this SHArP release 5 Mellanox Technologies Overview 1 Overview Mellanox Scalable Hierarchical Aggregation Protocol (SHArP™) technology improves upon the performance of MPI operation, by offloading collective operations from the CPU to the switch network, and eliminating the need to send data multiple times between endpoints. This innovative approach decreases the amount of data traversing the network as aggregation nodes are reached, and dramatically reduces the MPI operations time. Implementing collective communication algorithms in the network also has additional benefits, such as freeing up valuable CPU resources for computation rather than using them to process communication. 1.1 Supported OS and Platforms Distro Platform Kernel RHEL 6.1 x86-64 2.6.32-131.0.15 RHEL 6.2 x86-64 2.6.32-220 RHEL 6.3 x86-64 2.6.32-279 RHEL 6.4 x86-64 2.6.32-358 RHEL 6.5 x86-64 2.6.32-431 RHEL 7.0 x86-64 3.10.0-123 RHEL 7.2 x86-64 3.10.0-327 Fedora14 x86-64 2.6.35.6-45 Fedora16 x86-64 3.1.0-7 Fedora17 x86-64 3.3.4-5 Fedora18 x86-64 3.6.10-4 SLES 11 SP1 x86-64 2.6.32.12-0.7 SLES 11 SP2 x86-64 3.0.13-0.27 SLES 11 SP3 x86-64 3.0.76-0.11 Ubuntu12.04 x86-64 3.2.0-37 Ubuntu13.10 x86-64 3.11.0-12 Ubuntu14.4 x86-64 3.13.0-24 Ubuntu15.10 x86-64 4.2.0-16 CentOS6.3 x86-64 2.6.32-279 CentOS6.0 x86-64 2.6.32-71 6 Mellanox Technologies Scalable Hierarchical Aggregation Protocol (SHArP) Release Notes 2 Main Features in This Release SHArP 1.0.0 provides the following new features: Table 2: New Features, Changes and Fixes Category Description MPI 2.x Barrier and Allreduce collective operations using SHArP protocol are supported in Open MPI, MPICH, ScalableSHMEM with HCOLL library. HCOLL Enables running of SHArP collective with the mpirun utility. For the complete list of flags that can be used when running SHArP, please refer to the SHArP Deployment Guide. 7 Mellanox Technologies Limitations 3 Limitations Module Limitation Workaround Aggregation Manager Aggregation Manager High Availability protocol is currently not supported. Therefore, only a single instance of Aggregation Manager can run in the IB fabric. - Aggregation Manager Aggregation manager should run on the same Host where the Master Subnet Manager (SM) is running. - Aggregation Manager Upon Subnet Manager handover/failover, another instance of Aggregation Manager should be started on the Host where the new Master SM is running - Aggregation. Manager Upon restart of Aggregation Manager, ongoing jobs using SHArP collective operations may be terminated or get hanged. Postpone Aggregation Manager restart all jobs using SHArP collectives are completed. Aggregation Manager Aggregation Manager should be started after completion of fabric configuration by the Subnet Manager. - Aggregation Manager Switch reboot, fabric extension, and HCA cards replacement are currently not supported by the Aggregation Manager Restart Aggregation Manager after the Subnet Manager completes fabric reconfiguration followed by the fabric changes. Fabric Topology Only Fat-Tree and Quasi-Fat-Tree topologies are supported by the Aggregation Manager. - Fabric Topology Only IB fabrics where all compute nodes are connected to SHArP capable switches (SwitchIB-2) are supported by the Aggregation Manager. - Configuration Upon changes in configuration file, Aggregation Manager should be restarted to deploy new configuration. - SHArP Deamon Only x86 based servers platforms are currently supported. - 8 Mellanox Technologies