| Both sides previous revision Previous revision Next revision | Previous revision |
| dragnet:cluster_benchmark [2016-07-28 13:34] – [Infiniband] typo amesfoort | dragnet:cluster_benchmark [2017-09-19 21:40] (current) – [RDMA] fix typo amesfoort |
|---|
| ===== Infiniband ===== | ===== Infiniband ===== |
| |
| Each ''drgXX'' node has an FDR (54.545 Gbit/s) HCA (Host Channel Adapter) local to the second CPU (i.e. GPU 1). The 36 port cluster switch is connected to the COBALT switch with 5 aggregated lines (272.727 Gbit/s). See below what can be achieved under ideal circumstances. | Each ''drgXX'' node has an FDR (54.545 Gbit/s) HCA (Host Channel Adapter) local to the second CPU (i.e. CPU id 1). The 36 port cluster switch is connected to the COBALT switch with 5 aggregated lines (272.727 Gbit/s). See below what can be achieved under ideal circumstances. |
| |
| |
| ==== IPoIB: TCP and UDP ==== | ==== IPoIB: TCP and UDP ==== |
| |
| An application that uses the Infiniband (''ib'') network normally uses IPoIB (IP-over-Infiniband) to transfer data via TCP or UDP. DRAGNET IPoIB settings have been optimized for TCP (at the cost of UDP performance). We (mostly) use TCP and will not receive UDP data from LOFAR stations directly. | An application that uses the Infiniband (''ib'') network normally uses IPoIB (IP-over-Infiniband) to transfer data via TCP or UDP. DRAGNET IPoIB settings have been optimized for TCP (vs UDP performance) (IPoIB connected-mode enabled). We (mostly) use TCP and will not receive UDP data from LOFAR stations directly. |
| |
| We used the ''iperf3'' benchmark and got the following bandwidth numbers between two ''drgXX'' nodes: | We used the ''iperf3'' benchmark and got the following bandwidth numbers between two ''drgXX'' nodes: |
| RDMA (Remote Direct Memory Access) allows an application to directly access memory on another node. Although some initial administration is set up via the OS kernel, the actual transfer commands and completion handling does not go via the kernel. This also saves data copying on sender and receiver and CPU usage. | RDMA (Remote Direct Memory Access) allows an application to directly access memory on another node. Although some initial administration is set up via the OS kernel, the actual transfer commands and completion handling does not go via the kernel. This also saves data copying on sender and receiver and CPU usage. |
| |
| Typical applications that may use RDMA are applications that use MPI (Message Passing Interface) (such as COBALT), or (hopefully) the LUSTRE client. NFS can also be set up to use RDMA. You can program directly into the ''verbs'' and ''rdma-cm'' C APIs and link to those libraries, but be aware that extending some code to do this is not a 1 hr task... (Undoubtedly, there is also a Python module that either only wraps or even makes makes life easier.) | Typical applications that may use RDMA are applications that use MPI (Message Passing Interface) (such as COBALT), or (hopefully) the LUSTRE client. NFS can also be set up to use RDMA. You can program directly into the ''verbs'' and ''rdma-cm'' C APIs and link to those libraries, but be aware that extending some code to do this is not a 1 hr task... (Undoubtedly, there is also a Python module that either only wraps or even makes life easier.) |
| |
| We used the ''qperf'' benchmark and got the following bandwidth and latency numbers between two ''drgXX'' nodes (TCP/UDP/SCTP over IP also included, but not as fast as mentioned above): | We used the ''qperf'' benchmark and got the following bandwidth and latency numbers between two ''drgXX'' nodes (TCP/UDP/SCTP over IP also included, but not as fast as mentioned above): |
| ===== 10 Gbit/s Ethernet ===== | ===== 10 Gbit/s Ethernet ===== |
| |
| The ''drgXX'' and ''dragproc'' nodes have a 10 Gbit/s ethernet adapter local to the first CPU (i.e. GPU 0). A 48 port Ethernet switch is connected to a LOFAR core switch with 6 aggregated lines (60 Gbit/s). See below what can be achieved under ideal circumstances. | The ''drgXX'' and ''dragproc'' nodes have a 10 Gbit/s ethernet adapter local to the first CPU (i.e. CPU id 0). A 48 port Ethernet switch is connected to a LOFAR core switch with 6 aggregated lines (60 Gbit/s). See below what can be achieved under ideal circumstances. |
| |
| |
| ====== Storage ====== | ====== Storage ====== |
| |
| Storage numbers N/A | Only rough write tests have been done with a sequential dd(1). Disk I/O bandwidth changes across the platters. Actual file I/O also depends on how the filesystem lays out the data. |
| | |
| | |
| | ===== drgXX nodes ===== |
| | |
| | Scratch space on ''drgXX'' at ''/data1'' and ''/data2'' (individually). A transfer size of 4k vs 64k does not appear to matter. We reach up to 288 MiB/s. Copying a large file using cp(1) reaches 225 - 279 MiB/s. |
| | |
| | Another cp test on a 85+% full target filesystem: |
| | |
| | [amesfoort@drg23 data1]$ time (cp /data2/L412984/L412984_SAP000_B045_S0_P000_bf.raw /data2/L412984/L412984_SAP000_B046_S0_P000_bf.raw Ltmp && sync) |
| | |
| | real 9m56.566s |
| | user 0m0.799s |
| | sys 4m20.007s |
| | |
| | With a file size of 2 * 75931582464 bytes, that's a read/write speed of 242.8 MiB/s. |
| | ===== dragproc node ===== |
| | |
| | On ''dragproc'' at ''/data'', a transfer size of 64k may perform somewhat better than 4k, but not consistently. We reach 490 - 530 MiB/s. |
| |