Message-ID: <616f14aa-7b5f-a2ec-267d-1e5066502be2@sberdevices.ru>
Date:   Thu, 1 Dec 2022 11:46:10 +0000
From:   Arseniy Krasnov <AVKrasnov@...rdevices.ru>
To:     Stefano Garzarella <sgarzare@...hat.com>
CC:     "David S. Miller" <davem@...emloft.net>,
        "edumazet@...gle.com" <edumazet@...gle.com>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>,
        "kys@...rosoft.com" <kys@...rosoft.com>,
        "haiyangz@...rosoft.com" <haiyangz@...rosoft.com>,
        "wei.liu@...nel.org" <wei.liu@...nel.org>,
        Dexuan Cui <decui@...rosoft.com>,
        Bryan Tan <bryantan@...are.com>,
        Vishnu Dasa <vdasa@...are.com>,
        Krasnov Arseniy <oxffffaa@...il.com>,
        Bobby Eshleman <bobby.eshleman@...il.com>,
        "Bobby Eshleman" <bobby.eshleman@...edance.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "virtualization@...ts.linux-foundation.org" 
        <virtualization@...ts.linux-foundation.org>,
        "linux-hyperv@...r.kernel.org" <linux-hyperv@...r.kernel.org>,
        kernel <kernel@...rdevices.ru>
Subject: Re: [RFC PATCH v2 6/6] test/vsock: vsock_perf utility

On 01.12.2022 13:21, Stefano Garzarella wrote:
> On Fri, Nov 25, 2022 at 05:15:08PM +0000, Arseniy Krasnov wrote:
>> This adds utility to check vsock rx/tx performance.
>>
>> Usage as sender:
>> ./vsock_perf -p <port> -m <bytes to send)
>> Usage as receiver:
>> ./vsock_perf -c <cid> -p <port> -r <SO_RCVLOWAT>
>>
>> Signed-off-by: Arseniy Krasnov <AVKrasnov@...rdevices.ru>
>> ---
>> tools/testing/vsock/Makefile     |   1 +
>> tools/testing/vsock/README       |  34 +++
>> tools/testing/vsock/vsock_perf.c | 400 +++++++++++++++++++++++++++++++
>> 3 files changed, 435 insertions(+)
>> create mode 100644 tools/testing/vsock/vsock_perf.c
>>
>> diff --git a/tools/testing/vsock/Makefile b/tools/testing/vsock/Makefile
>> index f8293c6910c9..d36fdd59fe2e 100644
>> --- a/tools/testing/vsock/Makefile
>> +++ b/tools/testing/vsock/Makefile
>> @@ -3,6 +3,7 @@ all: test
>> test: vsock_test vsock_diag_test
>> vsock_test: vsock_test.o timeout.o control.o util.o
>> vsock_diag_test: vsock_diag_test.o timeout.o control.o util.o
>> +vsock_perf: vsock_perf.o
>>
>> CFLAGS += -g -O2 -Werror -Wall -I. -I../../include -I../../../usr/include -Wno-pointer-sign -fno-strict-overflow -fno-strict-aliasing -fno-common -MMD -U_FORTIFY_SOURCE -D_GNU_SOURCE
>> .PHONY: all test clean
>> diff --git a/tools/testing/vsock/README b/tools/testing/vsock/README
>> index 4d5045e7d2c3..dbeba1d62016 100644
>> --- a/tools/testing/vsock/README
>> +++ b/tools/testing/vsock/README
>> @@ -35,3 +35,37 @@ Invoke test binaries in both directions as follows:
>>                        --control-port=$GUEST_IP \
>>                        --control-port=1234 \
>>                        --peer-cid=3
> 
> I'm not a native speaker, so the suggestions below may be wrong.Ack, this and every comment below
Thanks for review!
Arseniy
> 
>> +
>> +vsock_perf utility
>> +-------------------
>> +'vsock_perf' is simple tool to measure vsock performance. It works in
> 
> is a simple tool
> 
>> +sender/receiver modes: sender waits for connection at specified port,
>> +and after accept, starts data transmission to receiver. After data
>   ^
> and after accepting it
>                                              ^
>                                              to the receiver
> 
>> +processing is done, it prints several metrics(see below).
>> +
>> +Usage:
>> +# run as sender
>> +# listen port 1234, tx buffer size is 1MB, send of data 1G
>> +./vsock_perf -s -p 1234 -b 1MB -m 1G
> 
> Like for vsock_test and vsock_diag_test I would use only the long form for the parameters (e.g. --sender, --port, etc.)
> 
>> +
>> +Output:
>> +tx performance: A Gb/s
>> +
>> +Output explanation:
>> +A is calculated as "number of bytes to send" / "time in tx loop"
>> +
>> +# run as receiver
>> +# connect to CID 2, port 1234, rx buffer size is 1MB, peer buf is 1G, SO_RCVLOWAT is 65536
>> +./vsock_perf -c 2 -p 1234 -b 1MB -v 1G -r 65536
>> +
>> +Output:
>> +rx performance: A Gb/s
>> +total in 'read()': B sec
>> +POLLIN wakeups: C
>> +average in 'read()': D ns
>> +
>> +Output explanation:
>> +A is calculated as "number of received bytes" / "time in rx loop".
>> +B is time, spent in 'read()' system call(excluding 'poll()')
>> +C is number of 'poll()' wake ups with POLLIN bit set.
>> +D is B / C, e.g. average amount of time, spent in single 'read()'.
>> diff --git a/tools/testing/vsock/vsock_perf.c b/tools/testing/vsock/vsock_perf.c
>> new file mode 100644
>> index 000000000000..bca2027de52d
>> --- /dev/null
>> +++ b/tools/testing/vsock/vsock_perf.c
>> @@ -0,0 +1,400 @@
>> +// SPDX-License-Identifier: GPL-2.0-only
>> +/*
>> + * vsock_perf - benchmark utility for vsock.
>> + *
>> + * Copyright (C) 2022 SberDevices.
>> + *
>> + * Author: Arseniy Krasnov <AVKrasnov@...rdevices.ru>
>> + */
>> +#include <getopt.h>
>> +#include <stdio.h>
>> +#include <stdlib.h>
>> +#include <stdbool.h>
>> +#include <string.h>
>> +#include <errno.h>
>> +#include <unistd.h>
>> +#include <time.h>
>> +#include <sys/mman.h>
>> +#include <stdint.h>
>> +#include <poll.h>
>> +#include <sys/socket.h>
>> +#include <linux/vm_sockets.h>
>> +
>> +#define DEFAULT_BUF_SIZE_BYTES    (128 * 1024)
>> +#define DEFAULT_TO_SEND_BYTES    (64 * 1024)
>> +#define DEFAULT_VSOCK_BUF_BYTES (256 * 1024)
>> +#define DEFAULT_RCVLOWAT_BYTES    1
>> +#define DEFAULT_PORT        1234
>> +#define DEFAULT_CID        2
>> +
>> +#define BYTES_PER_GB        (1024 * 1024 * 1024ULL)
>> +#define NSEC_PER_SEC        (1000000000ULL)
>> +
>> +static unsigned int port = DEFAULT_PORT;
>> +static unsigned long buf_size_bytes = DEFAULT_BUF_SIZE_BYTES;
>> +static unsigned long vsock_buf_bytes = DEFAULT_VSOCK_BUF_BYTES;
>> +
>> +static inline time_t current_nsec(void)
>> +{
>> +    struct timespec ts;
>> +
>> +    if (clock_gettime(CLOCK_REALTIME, &ts)) {
>> +        perror("clock_gettime");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    return (ts.tv_sec * NSEC_PER_SEC) + ts.tv_nsec;
>> +}
>> +
>> +/* From lib/cmdline.c. */
>> +static unsigned long memparse(const char *ptr)
>> +{
>> +    char *endptr;
>> +
>> +    unsigned long long ret = strtoull(ptr, &endptr, 0);
>> +
>> +    switch (*endptr) {
>> +    case 'E':
>> +    case 'e':
>> +        ret <<= 10;
>> +    case 'P':
>> +    case 'p':
>> +        ret <<= 10;
>> +    case 'T':
>> +    case 't':
>> +        ret <<= 10;
>> +    case 'G':
>> +    case 'g':
>> +        ret <<= 10;
>> +    case 'M':
>> +    case 'm':
>> +        ret <<= 10;
>> +    case 'K':
>> +    case 'k':
>> +        ret <<= 10;
>> +        endptr++;
>> +    default:
>> +        break;
>> +    }
>> +
>> +    return ret;
>> +}
>> +
>> +static void vsock_increase_buf_size(int fd)
>> +{
>> +    if (setsockopt(fd, AF_VSOCK, SO_VM_SOCKETS_BUFFER_MAX_SIZE,
>> +               &vsock_buf_bytes, sizeof(vsock_buf_bytes))) {
>> +        perror("setsockopt");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    if (setsockopt(fd, AF_VSOCK, SO_VM_SOCKETS_BUFFER_SIZE,
>> +               &vsock_buf_bytes, sizeof(vsock_buf_bytes))) {
>> +        perror("setsockopt");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +}
>> +
>> +static int vsock_connect(unsigned int cid, unsigned int port)
>> +{
>> +    union {
>> +        struct sockaddr sa;
>> +        struct sockaddr_vm svm;
>> +    } addr = {
>> +        .svm = {
>> +            .svm_family = AF_VSOCK,
>> +            .svm_port = port,
>> +            .svm_cid = cid,
>> +        },
>> +    };
>> +    int fd;
>> +
>> +    fd = socket(AF_VSOCK, SOCK_STREAM, 0);
>> +
>> +    if (fd < 0)
>> +        return -1;
>> +
>> +    vsock_increase_buf_size(fd);
>> +
>> +    if (connect(fd, &addr.sa, sizeof(addr.svm)) < 0) {
>> +        close(fd);
>> +        return -1;
>> +    }
>> +
>> +    return fd;
>> +}
>> +
>> +static float get_gbps(unsigned long bytes, time_t ns_delta)
>> +{
>> +    return ((float)bytes / BYTES_PER_GB) /
>> +           ((float)ns_delta / NSEC_PER_SEC);
>> +}
>> +
>> +static void run_sender(unsigned long to_send_bytes)
>> +{
>> +    time_t tx_begin_ns;
>> +    size_t total_send;
>> +    int client_fd;
>> +    char *data;
>> +    int fd;
>> +    union {
>> +        struct sockaddr sa;
>> +        struct sockaddr_vm svm;
>> +    } addr = {
>> +        .svm = {
>> +            .svm_family = AF_VSOCK,
>> +            .svm_port = port,
>> +            .svm_cid = VMADDR_CID_ANY,
>> +        },
>> +    };
>> +    union {
>> +        struct sockaddr sa;
>> +        struct sockaddr_vm svm;
>> +    } clientaddr;
>> +
>> +    socklen_t clientaddr_len = sizeof(clientaddr.svm);
>> +
>> +    printf("Run as sender\n");
>> +    printf("Listen port %u\n", port);
>> +    printf("Send %lu bytes\n", to_send_bytes);
>> +    printf("TX buffer %lu bytes\n", buf_size_bytes);
>> +    printf("Peer buffer %lu bytes\n", vsock_buf_bytes);
>> +
>> +    fd = socket(AF_VSOCK, SOCK_STREAM, 0);
>> +
>> +    if (fd < 0) {
>> +        perror("socket");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    if (bind(fd, &addr.sa, sizeof(addr.svm)) < 0) {
>> +        perror("bind");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    if (listen(fd, 1) < 0) {
>> +        perror("listen");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    client_fd = accept(fd, &clientaddr.sa, &clientaddr_len);
>> +
>> +    if (client_fd < 0) {
>> +        perror("accept");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    vsock_increase_buf_size(client_fd);
>> +
>> +    data = malloc(buf_size_bytes);
>> +
>> +    if (!data) {
>> +        printf("malloc failed\n");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    memset(data, 0, buf_size_bytes);
>> +    total_send = 0;
>> +    tx_begin_ns = current_nsec();
>> +
>> +    while (total_send < to_send_bytes) {
>> +        ssize_t sent;
>> +
>> +        sent = write(client_fd, data, buf_size_bytes);
>> +
>> +        if (sent <= 0) {
>> +            perror("write");
>> +            exit(EXIT_FAILURE);
>> +        }
>> +
>> +        total_send += sent;
>> +    }
>> +
>> +    printf("tx performance: %f Gb/s\n",
>> +           get_gbps(total_send, current_nsec() - tx_begin_ns));
>> +
>> +    close(client_fd);
>> +    close(fd);
>> +
>> +    free(data);
>> +}
>> +
>> +static void run_receiver(int peer_cid, unsigned long rcvlowat_bytes)
>> +{
>> +    unsigned int read_cnt;
>> +    time_t rx_begin_ns;
>> +    time_t in_read_ns;
>> +    size_t total_recv;
>> +    void *data;
>> +    int fd;
>> +
>> +    printf("Run as receiver\n");
>> +    printf("Connect to %i:%u\n", peer_cid, port);
>> +    printf("RX buffer %lu bytes\n", buf_size_bytes);
>> +    printf("Peer buffer %lu bytes\n", vsock_buf_bytes);
>> +    printf("SO_RCVLOWAT %lu bytes\n", rcvlowat_bytes);
>> +
>> +    fd = vsock_connect(peer_cid, port);
>> +
>> +    if (fd < 0) {
>> +        perror("socket");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    if (setsockopt(fd, SOL_SOCKET, SO_RCVLOWAT,
>> +               &rcvlowat_bytes,
>> +               sizeof(rcvlowat_bytes))) {
>> +        perror("setsockopt");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    data = mmap(NULL, buf_size_bytes, PROT_READ | PROT_WRITE,
>> +            MAP_ANONYMOUS | MAP_PRIVATE | MAP_POPULATE, -1, 0);
>> +
>> +    if (data == MAP_FAILED) {
>> +        perror("mmap");
>> +        exit(EXIT_FAILURE);
>> +    }
>> +
>> +    read_cnt = 0;
>> +    in_read_ns = 0;
>> +    total_recv = 0;
>> +    rx_begin_ns = current_nsec();
>> +
>> +    while (1) {
>> +        struct pollfd fds = { 0 };
>> +
>> +        fds.fd = fd;
>> +        fds.events = POLLIN | POLLERR | POLLHUP |
>> +                 POLLRDHUP | POLLNVAL;
>> +
>> +        if (poll(&fds, 1, -1) < 0) {
>> +            perror("poll");
>> +            exit(EXIT_FAILURE);
>> +        }
>> +
>> +        if (fds.revents & POLLERR) {
>> +            printf("'poll()' error\n");
>> +            exit(EXIT_FAILURE);
>> +        }
>> +
>> +        if (fds.revents & POLLIN) {
>> +            ssize_t bytes_read;
>> +            time_t t;
>> +
>> +            t = current_nsec();
>> +            bytes_read = read(fd, data, buf_size_bytes);
>> +            in_read_ns += (current_nsec() - t);
>> +            read_cnt++;
>> +
>> +            if (!bytes_read)
>> +                break;
>> +
>> +            if (bytes_read < 0) {
>> +                perror("recv");
>> +                exit(EXIT_FAILURE);
>> +            }
>> +
>> +            total_recv += bytes_read;
>> +        }
>> +
>> +        if (fds.revents & (POLLHUP | POLLRDHUP))
>> +            break;
>> +    }
>> +
>> +    printf("rx performance: %f Gb/s\n",
>> +           get_gbps(total_recv, current_nsec() - rx_begin_ns));
>> +    printf("total in 'read()': %f sec\n", (float)in_read_ns / NSEC_PER_SEC);
> 
> Maybe better "total time in ..."
> 
>> +    printf("POLLIN wakeups: %i\n", read_cnt);
>> +    printf("average in 'read()': %f ns\n", (float)in_read_ns / read_cnt);
> 
> Ditto.
> 
>> +
>> +    munmap(data, buf_size_bytes);
>> +    close(fd);
>> +}
>> +
>> +static void usage(void)
>> +{
>> +    printf("Help:\n"
>> +           "\n"
>> +           "This is benchmarking utility, to test vsock performance.\n"
>> +           "It runs in two modes: sender or receiver. In sender mode, it waits\n"
>> +           "connection from receiver, and when established, sender starts data\n"
>> +           "transmission.\n"
>> +           "\n"
>> +           "Options:\n"
>> +           "  -h                This help message\n"
>> +           "  -s                Sender mode(receiver default)\n"
>> +           "  -p <port>            Port (%d)\n"
>> +           "  -c <cid>            CID of the peer (%d)\n"
>> +           "  -m <bytes to send>        Bytes to send (%d)\n"
>> +           "  -b <buffer size>        Rx/Tx buffer size (%d). In sender mode\n"
>> +           "                                it is size of buffer passed to 'write()'.\n"
>> +           "                                In receiver mode it is size of buffer passed\n"
>> +           "                                to 'read()'.\n"
>> +           "  -v <peer buffer size>        Peer buffer size (%d)\n"
> 
> Instead of "peer buffer", what about "socket buffer"?
> 
>> +           "  -r <SO_RCVLOWAT>        SO_RCVLOWAT (%d)\n"
>> +           "\n", DEFAULT_PORT, DEFAULT_CID, DEFAULT_TO_SEND_BYTES,
>> +           DEFAULT_BUF_SIZE_BYTES, DEFAULT_VSOCK_BUF_BYTES,
>> +           DEFAULT_RCVLOWAT_BYTES);
>> +    exit(EXIT_FAILURE);
>> +}
>> +
>> +static long strtolx(const char *arg)
>> +{
>> +    long value;
>> +    char *end;
>> +
>> +    value = strtol(arg, &end, 10);
>> +
>> +    if (end != arg + strlen(arg))
>> +        usage();
>> +
>> +    return value;
>> +}
>> +
>> +int main(int argc, char **argv)
>> +{
>> +    unsigned long to_send_bytes = DEFAULT_TO_SEND_BYTES;
>> +    unsigned long rcvlowat_bytes = DEFAULT_RCVLOWAT_BYTES;
>> +    bool receiver_mode = true;
>> +    int peer_cid = DEFAULT_CID;
>> +    int c;
>> +
>> +    while ((c = getopt(argc, argv, "v:r:c:p:m:b:sh")) != -1) {
>> +        switch (c) {
>> +        case 'v': /* Peer buffer size. */
>> +            vsock_buf_bytes = memparse(optarg);
>> +            break;
>> +        case 'r': /* SO_RCVLOWAT value. */
>> +            rcvlowat_bytes = memparse(optarg);
>> +            break;
>> +        case 'c': /* CID to connect to. */
>> +            peer_cid = strtolx(optarg);
>> +            break;
>> +        case 'p': /* Port to connect to. */
>> +            port = strtolx(optarg);
>> +            break;
>> +        case 'm': /* Bytes to send. */
>> +            to_send_bytes = memparse(optarg);
>> +            break;
>> +        case 'b': /* Size of rx/tx buffer. */
>> +            buf_size_bytes = memparse(optarg);
>> +            break;
>> +        case 's': /* Sender mode. */
>> +            receiver_mode = false;
>> +            break;
>> +        case 'h': /* Help. */
>> +            usage();
>> +            break;
>> +        default:
>> +            usage();
>> +        }
>> +    }
>> +
>> +    if (receiver_mode)
>> +        run_receiver(peer_cid, rcvlowat_bytes);
>> +    else
>> +        run_sender(to_send_bytes);
>> +
>> +    return 0;
>> +}
>> -- 
>> 2.25.1
> 

Powered by blists - more mailing lists