Skip to content

Commit cce1aeb

Browse files
GitHubber17GitHubber17
authored
Merge pull request #129470 from vermagit/patch-6
Plots for lat/bw
2 parents 838386e + 71e18c1 commit cce1aeb

File tree

2 files changed

+30
-77
lines changed

2 files changed

+30
-77
lines changed

‎articles/virtual-machines/workloads/hpc/hc-series-performance.md

Lines changed: 26 additions & 71 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
---
2-
title: HC-series VM size performance - Azure Virtual Machines | Microsoft Docs
2+
title: HC-series VM size performance
33
description: Learn about performance testing results for HC-series VM sizes in Azure.
44
services: virtual-machines
55
documentationcenter: ''
@@ -11,7 +11,7 @@ tags: azure-resource-manager
1111
ms.service: virtual-machines
1212
ms.workload: infrastructure-services
1313
ms.topic: article
14-
ms.date: 05/15/2019
14+
ms.date: 09/10/2020
1515
ms.author: amverma
1616
---
1717

@@ -21,87 +21,42 @@ Several performance tests have been run on HC-series sizes. The following are so
2121

2222
| Workload | HB |
2323
|-------------------------------------------------|-----------------------|
24-
| STREAM Triad | ~190 GB/s (Intel MLC AVX-512) |
25-
| High-Performance Linpack (HPL) | ~3520 GigaFLOPS (Rpeak), ~2970 GigaFLOPS (Rmax) |
26-
| RDMA latency & bandwidth | 1.80 microseconds, 96.3 Gb/s |
27-
| FIO on local NVMe SSD | ~1.3 GB/s reads, ~900 MB/s writes |
28-
| IOR on 4 Azure Premium SSD (P30 Managed Disks, RAID0)** | ~780 MB/s reads, ~780 MB/writes |
24+
| STREAM Triad | 190 GB/s (Intel MLC AVX-512) |
25+
| High-Performance Linpack (HPL) | 3520 GigaFLOPS (Rpeak), 2970 GigaFLOPS (Rmax) |
26+
| RDMA latency & bandwidth | 1.05 microseconds, 96.8 Gb/s |
27+
| FIO on local NVMe SSD | 1.3 GB/s reads, 900 MB/s writes |
28+
| IOR on 4 Azure Premium SSD (P30 Managed Disks, RAID0)** | 780 MB/s reads, 780 MB/writes |
2929

30-
## InfiniBand send latency
30+
## MPI latency
3131

32-
Mellanox Perftest.
33-
34-
```bash
35-
numactl --physcpubind=[INSERT CORE #] ib_send_lat -a
36-
```
37-
38-
| #bytes | #iterations | t_min[microsecond] | t_max[microsecond] | t_typical[microsecond] | t_avg[microsecond] | t_stdev[microsecond] |
39-
|-----------------|-----------------|-----------------|-----------------|-----------------|-----------------|-----------------|
40-
| 2 | 1000 | 1.80 | 7.50 | 1.85 | 1.86 | 0.20 |
41-
| 4 | 1000 | 1.79 | 6.06 | 1.83 | 1.84 | 0.20 |
42-
| 8 | 1000 | 1.78 | 5.26 | 1.83 | 1.84 | 0.19 |
43-
| 16 | 1000 | 1.79 | 6.21 | 1.83 | 1.84 | 0.22 |
44-
| 32 | 1000 | 1.80 | 6.82 | 1.84 | 1.85 | 0.24 |
45-
| 64 | 1000 | 1.85 | 5.47 | 1.88 | 1.86 | 0.12 |
46-
| 128 | 1000 | 1.88 | 5.61 | 1.93 | 1.89 | 0.25 |
47-
| 256 | 1000 | 2.24 | 6.39 | 2.28 | 2.02 | 0.18 |
48-
| 512 | 1000 | 2.32 | 5.42 | 2.36 | 2.30 | 0.17 |
49-
| 1024 | 1000 | 2.43 | 6.22 | 2.48 | 2.38 | 0.21 |
50-
| 2048 | 1000 | 2.68 | 6.14 | 2.75 | 2.52 | 0.20 |
51-
| 4096 | 1000 | 3.17 | 7.02 | 3.26 | 2.81 | 0.24 |
52-
53-
## OSU MPI Latency Test
54-
55-
OSU MPI Latency Test v5.4.3.
32+
MPI latency test from the OSU microbenchmark suite is run. Sample scripts are on [GitHub](https://github.com/Azure/azhpc-images/blob/04ddb645314a6b2b02e9edb1ea52f079241f1297/tests/run-tests.sh)
5633

5734
```bash
5835
./bin/mpirun_rsh -np 2 -hostfile ~/hostfile MV2_CPU_MAPPING=[INSERT CORE #] ./osu_latency
5936
```
6037
61-
| #bytes | Latency [microsecond] (MPICH 3.3 + CH4) | Latency [microsecond] (OpenMPI 4.0.0) | Latency [microsecond] (MVAPICH2 2.3) |
62-
|------|----------|----------|----------|
63-
| 2 | 1.84 | 1.78 | 2.08 |
64-
| 4 | 1.84 | 1.79 | 2.08 |
65-
| 8 | 1.85 | 1.79 | 2.05 |
66-
| 16 | 1.85 | 1.79 | 2.1 |
67-
| 32 | 1.87 | 1.82 | 2.12 |
68-
| 64 | 2 | 1.95 | 2.13 |
69-
| 128 | 2.05 | 2 | 2.18 |
70-
| 256 | 2.48 | 2.44 | 2.75 |
71-
| 512 | 2.57 | 2.52 | 2.81 |
72-
| 1024 | 2.76 | 2.71 | 2.97 |
73-
| 2048 | 3.09 | 3.11 | 3.34 |
74-
| 4096 | 3.72 | 3.91 | 4.44 |
38+
:::image type="content" source="./media/latency-hc.png" alt-text="MPI latency on Azure HC.":::
7539
7640
## MPI bandwidth
7741
78-
OSU MPI Bandwidth Test v5.4.3.
42+
MPI bandwidth test from the OSU microbenchmark suite is run. Sample scripts are on [GitHub](https://github.com/Azure/azhpc-images/blob/04ddb645314a6b2b02e9edb1ea52f079241f1297/tests/run-tests.sh)
7943
8044
```bash
8145
./mvapich2-2.3.install/bin/mpirun_rsh -np 2 -hostfile ~/hostfile MV2_CPU_MAPPING=[INSERT CORE #] ./mvapich2-2.3/osu_benchmarks/mpi/pt2pt/osu_bw
8246
```
8347
84-
| #Size | Bandwidth (MB/s) | Bandwidth (Gb/s) |
85-
|---------|------------------|------------------|
86-
| 2 | 6.18 | 0.04944 |
87-
| 4 | 13.27 | 0.10616 |
88-
| 8 | 26.58 | 0.21264 |
89-
| 16 | 53.51 | 0.42808 |
90-
| 32 | 106.81 | 0.85448 |
91-
| 64 | 211.24 | 1.68992 |
92-
| 128 | 386.98 | 3.09584 |
93-
| 256 | 756.32 | 6.05056 |
94-
| 512 | 1434.2 | 11.4736 |
95-
| 1024 | 2663.8 | 21.3104 |
96-
| 2048 | 4396.99 | 35.17592 |
97-
| 4096 | 6365.86 | 50.92688 |
98-
| 8192 | 8137.9 | 65.1032 |
99-
| 16384 | 9218.29 | 73.74632 |
100-
| 32768 | 10564.61 | 84.51688 |
101-
| 65536 | 11275.6 | 90.2048 |
102-
| 131072 | 11633.7 | 93.0696 |
103-
| 262144 | 11856.27 | 94.85016 |
104-
| 524288 | 11962.69 | 95.70152 |
105-
| 1048576 | 12025.43 | 96.20344 |
106-
| 2097152 | 12038.7 | 96.3096 |
107-
| 4194304 | 11290.92 | 90.32736 |
48+
:::image type="content" source="./media/bandwidth-hc.png" alt-text="MPI bandwidth on Azure HC.":::
49+
50+
51+
## Mellanox Perftest
52+
53+
The [Mellanox Perftest package](https://community.mellanox.com/s/article/perftest-package) has many InfiniBand tests such as latency (ib_send_lat) and bandwidth (ib_send_bw). An example command is below.
54+
55+
```console
56+
numactl --physcpubind=[INSERT CORE #] ib_send_lat -a
57+
```
58+
59+
## Next steps
60+
61+
- Read about the latest announcements and some High Performance Computing (HPC) examples and results at the [Azure Compute Tech Community Blogs](https://techcommunity.microsoft.com/t5/azure-compute/bg-p/AzureCompute).
62+
- For a higher-level architectural view of running HPC workloads, see [High Performance Computing (HPC) on Azure](/azure/architecture/topics/high-performance-computing/).

‎docfx.json

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -783,12 +783,10 @@
783783
"articles/app-service/scripts/*.yml": "Azure App Service",
784784
"articles/app-service/environment/*.md": "Azure App Service Environment",
785785
"articles/app-service/environment/*.yml": "Azure App Service Environment",
786-
"articles/virtual-machines/*.md": "Azure Virtual Machines",
787-
"articles/virtual-machines/*.yml": "Azure Virtual Machines",
788-
"articles/virtual-machines/linux/*.md": "Azure Linux Virtual Machines",
789-
"articles/virtual-machines/linux/*.yml": "Azure Linux Virtual Machines",
790-
"articles/virtual-machines/windows/*.md": "Azure Windows Virtual Machines",
791-
"articles/virtual-machines/windows/*.yml": "Azure Windows Virtual Machines",
786+
"articles/virtual-machines/**/*.md": "Azure Virtual Machines",
787+
"articles/virtual-machines/**/*.yml": "Azure Virtual Machines",
788+
"articles/virtual-machines/workloads/**/*.md": "Azure Virtual Machines",
789+
"articles/virtual-machines/workloads/**/*.yml": "Azure Virtual Machines",
792790
"articles/virtual-machine-scale-sets/*.md": "Azure Virtual Machine Scale Sets",
793791
"articles/virtual-machine-scale-sets/*.yml": "Azure Virtual Machine Scale Sets",
794792
"articles/batch/**/*.md": "Azure Batch",

0 commit comments

Comments
 (0)