http2

README

http2-bench

Setup

1. Install nghttp2. If you have apt-get installed, ./setup_nghttp.sh will do all the required setup.
2. Start AIStore with any number of proxies and one target. Since nghttp2 does not follow redirects, the benchmarking scripts must point directly at a target.

Usage

Each benchmark has four arguments:

• U: The URL of the target

• B: The bucket to get files from

• N: The number of files to get

• C: The number of concurrent clients to use (unavailable for server_push_http2_bench)

• ./server_push_http2_bench U B N uses http2 with server push to get N files.

• ./no_server_push_http2_bench U B N C uses http2 to get N files, with C clients.

• ./http1_bench U B N C uses http1.1 to get N files, with C clients.

Examples

$ ./server_push_http2_bench localhost:8081 local_benchmark_bucket 100
$ ./no_server_push_http2_bench localhost:8081 local_benchmark_bucket 100 5
$ ./http1_bench localhost:8081 local_benchmark_bucket 100 5

Helpers

1. put_files.go puts a given number of files with a given size into AIStore. It has five command-line parameters:

• -url: the URL of the AIStore proxy (default: http://localhost:8080)
• -files: the number of files to put (default: 10)
• -filesize: the size of files to put, in KB (default: 1)
• -bucket: the name of the bucket to put them in (default: local_benchmark_bucket)
• -workers: The number of workers to use to put files (default: 10)

1. create_local_bucket creates an ais bucket in AIStore. It takes one parameter:

• B: The ais bucket to create

Examples

$ ./create_local_bucket local_benchmark_bucket
$ go run put_files.go -files 80 -filesize 50 -bucket local_benchmark_bucket -workers 30

Limitations

• Read chunk size cannot be modified
• Redirects are not followed by nghttp2, so the benchmark must be pointed directly to one target
• server_push_http2_bench uses only one concurrent client

Documentation

Overview

Package http2 puts a given number of files with a given size into AIStore. See /bench/http2/README.md for more info.

• Copyright (c) 2018-2021, NVIDIA CORPORATION. All rights reserved. *