2019-03-01 13:38:49 -07:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
/* Copyright (c) 2019 Facebook
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of version 2 of the GNU General Public
|
|
|
|
* License as published by the Free Software Foundation.
|
|
|
|
*
|
|
|
|
* Example program for Host Bandwidth Managment
|
|
|
|
*
|
|
|
|
* This program loads a cgroup skb BPF program to enforce cgroup output
|
|
|
|
* (egress) or input (ingress) bandwidth limits.
|
|
|
|
*
|
|
|
|
* USAGE: hbm [-d] [-l] [-n <id>] [-r <rate>] [-s] [-t <secs>] [-w] [-h] [prog]
|
|
|
|
* Where:
|
|
|
|
* -d Print BPF trace debug buffer
|
|
|
|
* -l Also limit flows doing loopback
|
|
|
|
* -n <#> To create cgroup \"/hbm#\" and attach prog
|
|
|
|
* Default is /hbm1
|
2019-05-28 17:59:39 -06:00
|
|
|
* --no_cn Do not return cn notifications
|
2019-03-01 13:38:49 -07:00
|
|
|
* -r <rate> Rate limit in Mbps
|
|
|
|
* -s Get HBM stats (marked, dropped, etc.)
|
2019-03-05 10:31:13 -07:00
|
|
|
* -t <time> Exit after specified seconds (default is 0)
|
2019-03-01 13:38:49 -07:00
|
|
|
* -w Work conserving flag. cgroup can increase its bandwidth
|
|
|
|
* beyond the rate limit specified while there is available
|
|
|
|
* bandwidth. Current implementation assumes there is only
|
|
|
|
* NIC (eth0), but can be extended to support multiple NICs.
|
|
|
|
* Currrently only supported for egress.
|
|
|
|
* -h Print this info
|
|
|
|
* prog BPF program file name. Name defaults to hbm_out_kern.o
|
|
|
|
*/
|
|
|
|
|
|
|
|
#define _GNU_SOURCE
|
|
|
|
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <assert.h>
|
|
|
|
#include <sys/resource.h>
|
|
|
|
#include <sys/time.h>
|
|
|
|
#include <unistd.h>
|
|
|
|
#include <errno.h>
|
|
|
|
#include <fcntl.h>
|
|
|
|
#include <linux/unistd.h>
|
|
|
|
|
|
|
|
#include <linux/bpf.h>
|
|
|
|
#include <bpf/bpf.h>
|
2019-05-28 17:59:39 -06:00
|
|
|
#include <getopt.h>
|
2019-03-01 13:38:49 -07:00
|
|
|
|
|
|
|
#include "bpf_load.h"
|
|
|
|
#include "bpf_rlimit.h"
|
|
|
|
#include "cgroup_helpers.h"
|
|
|
|
#include "hbm.h"
|
|
|
|
#include "bpf_util.h"
|
2020-01-20 06:06:49 -07:00
|
|
|
#include <bpf/bpf.h>
|
|
|
|
#include <bpf/libbpf.h>
|
2019-03-01 13:38:49 -07:00
|
|
|
|
|
|
|
bool outFlag = true;
|
|
|
|
int minRate = 1000; /* cgroup rate limit in Mbps */
|
|
|
|
int rate = 1000; /* can grow if rate conserving is enabled */
|
|
|
|
int dur = 1;
|
|
|
|
bool stats_flag;
|
|
|
|
bool loopback_flag;
|
|
|
|
bool debugFlag;
|
|
|
|
bool work_conserving_flag;
|
2019-05-28 17:59:39 -06:00
|
|
|
bool no_cn_flag;
|
bpf: Add support for fq's EDT to HBM
Adds support for fq's Earliest Departure Time to HBM (Host Bandwidth
Manager). Includes a new BPF program supporting EDT, and also updates
corresponding programs.
It will drop packets with an EDT of more than 500us in the future
unless the packet belongs to a flow with less than 2 packets in flight.
This is done so each flow has at least 2 packets in flight, so they
will not starve, and also to help prevent delayed ACK timeouts.
It will also work with ECN enabled traffic, where the packets will be
CE marked if their EDT is more than 50us in the future.
The table below shows some performance numbers. The flows are back to
back RPCS. One server sending to another, either 2 or 4 flows.
One flow is a 10KB RPC, the rest are 1MB RPCs. When there are more
than one flow of a given RPC size, the numbers represent averages.
The rate limit applies to all flows (they are in the same cgroup).
Tests ending with "-edt" ran with the new BPF program supporting EDT.
Tests ending with "-hbt" ran on top HBT qdisc with the specified rate
(i.e. no HBM). The other tests ran with the HBM BPF program included
in the HBM patch-set.
EDT has limited value when using DCTCP, but it helps in many cases when
using Cubic. It usually achieves larger link utilization and lower
99% latencies for the 1MB RPCs.
HBM ends up queueing a lot of packets with its default parameter values,
reducing the goodput of the 10KB RPCs and increasing their latency. Also,
the RTTs seen by the flows are quite large.
Aggr 10K 10K 10K 1MB 1MB 1MB
Limit rate drops RTT rate P90 P99 rate P90 P99
Test rate Flows Mbps % us Mbps us us Mbps ms ms
-------- ---- ----- ---- ----- --- ---- ---- ---- ---- ---- ----
cubic 1G 2 904 0.02 108 257 511 539 647 13.4 24.5
cubic-edt 1G 2 982 0.01 156 239 656 967 743 14.0 17.2
dctcp 1G 2 977 0.00 105 324 408 744 653 14.5 15.9
dctcp-edt 1G 2 981 0.01 142 321 417 811 660 15.7 17.0
cubic-htb 1G 2 919 0.00 1825 40 2822 4140 879 9.7 9.9
cubic 200M 2 155 0.30 220 81 532 655 74 283 450
cubic-edt 200M 2 188 0.02 222 87 1035 1095 101 84 85
dctcp 200M 2 188 0.03 111 77 912 939 111 76 325
dctcp-edt 200M 2 188 0.03 217 74 1416 1738 114 76 79
cubic-htb 200M 2 188 0.00 5015 8 14ms 15ms 180 48 50
cubic 1G 4 952 0.03 110 165 516 546 262 38 154
cubic-edt 1G 4 973 0.01 190 111 1034 1314 287 65 79
dctcp 1G 4 951 0.00 103 180 617 905 257 37 38
dctcp-edt 1G 4 967 0.00 163 151 732 1126 272 43 55
cubic-htb 1G 4 914 0.00 3249 13 7ms 8ms 300 29 34
cubic 5G 4 4236 0.00 134 305 490 624 1310 10 17
cubic-edt 5G 4 4865 0.00 156 306 425 759 1520 10 16
dctcp 5G 4 4936 0.00 128 485 221 409 1484 7 9
dctcp-edt 5G 4 4924 0.00 148 390 392 623 1508 11 26
v1 -> v2: Incorporated Andrii's suggestions
v2 -> v3: Incorporated Yonghong's suggestions
v3 -> v4: Removed credit update that is not needed
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-07-02 16:09:52 -06:00
|
|
|
bool edt_flag;
|
2019-03-01 13:38:49 -07:00
|
|
|
|
|
|
|
static void Usage(void);
|
|
|
|
static void read_trace_pipe2(void);
|
|
|
|
static void do_error(char *msg, bool errno_flag);
|
|
|
|
|
|
|
|
#define DEBUGFS "/sys/kernel/debug/tracing/"
|
|
|
|
|
|
|
|
struct bpf_object *obj;
|
|
|
|
int bpfprog_fd;
|
|
|
|
int cgroup_storage_fd;
|
|
|
|
|
|
|
|
static void read_trace_pipe2(void)
|
|
|
|
{
|
|
|
|
int trace_fd;
|
|
|
|
FILE *outf;
|
|
|
|
char *outFname = "hbm_out.log";
|
|
|
|
|
|
|
|
trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
|
|
|
|
if (trace_fd < 0) {
|
|
|
|
printf("Error opening trace_pipe\n");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Future support of ingress
|
|
|
|
// if (!outFlag)
|
|
|
|
// outFname = "hbm_in.log";
|
|
|
|
outf = fopen(outFname, "w");
|
|
|
|
|
|
|
|
if (outf == NULL)
|
|
|
|
printf("Error creating %s\n", outFname);
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
static char buf[4097];
|
|
|
|
ssize_t sz;
|
|
|
|
|
|
|
|
sz = read(trace_fd, buf, sizeof(buf) - 1);
|
|
|
|
if (sz > 0) {
|
|
|
|
buf[sz] = 0;
|
|
|
|
puts(buf);
|
|
|
|
if (outf != NULL) {
|
|
|
|
fprintf(outf, "%s\n", buf);
|
|
|
|
fflush(outf);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void do_error(char *msg, bool errno_flag)
|
|
|
|
{
|
|
|
|
if (errno_flag)
|
|
|
|
printf("ERROR: %s, errno: %d\n", msg, errno);
|
|
|
|
else
|
|
|
|
printf("ERROR: %s\n", msg);
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int prog_load(char *prog)
|
|
|
|
{
|
|
|
|
struct bpf_prog_load_attr prog_load_attr = {
|
|
|
|
.prog_type = BPF_PROG_TYPE_CGROUP_SKB,
|
|
|
|
.file = prog,
|
|
|
|
.expected_attach_type = BPF_CGROUP_INET_EGRESS,
|
|
|
|
};
|
|
|
|
int map_fd;
|
|
|
|
struct bpf_map *map;
|
|
|
|
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (access(prog, O_RDONLY) < 0) {
|
|
|
|
printf("Error accessing file %s: %s\n", prog, strerror(errno));
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
if (bpf_prog_load_xattr(&prog_load_attr, &obj, &bpfprog_fd))
|
|
|
|
ret = 1;
|
|
|
|
if (!ret) {
|
|
|
|
map = bpf_object__find_map_by_name(obj, "queue_stats");
|
|
|
|
map_fd = bpf_map__fd(map);
|
|
|
|
if (map_fd < 0) {
|
|
|
|
printf("Map not found: %s\n", strerror(map_fd));
|
|
|
|
ret = 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ret) {
|
2019-11-06 17:51:52 -07:00
|
|
|
printf("ERROR: bpf_prog_load_xattr failed for: %s\n", prog);
|
2019-03-01 13:38:49 -07:00
|
|
|
printf(" Output from verifier:\n%s\n------\n", bpf_log_buf);
|
|
|
|
ret = -1;
|
|
|
|
} else {
|
|
|
|
ret = map_fd;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int run_bpf_prog(char *prog, int cg_id)
|
|
|
|
{
|
|
|
|
int map_fd;
|
|
|
|
int rc = 0;
|
|
|
|
int key = 0;
|
|
|
|
int cg1 = 0;
|
|
|
|
int type = BPF_CGROUP_INET_EGRESS;
|
|
|
|
char cg_dir[100];
|
|
|
|
struct hbm_queue_stats qstats = {0};
|
|
|
|
|
|
|
|
sprintf(cg_dir, "/hbm%d", cg_id);
|
|
|
|
map_fd = prog_load(prog);
|
|
|
|
if (map_fd == -1)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
if (setup_cgroup_environment()) {
|
|
|
|
printf("ERROR: setting cgroup environment\n");
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
cg1 = create_and_get_cgroup(cg_dir);
|
|
|
|
if (!cg1) {
|
|
|
|
printf("ERROR: create_and_get_cgroup\n");
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
if (join_cgroup(cg_dir)) {
|
|
|
|
printf("ERROR: join_cgroup\n");
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
qstats.rate = rate;
|
|
|
|
qstats.stats = stats_flag ? 1 : 0;
|
|
|
|
qstats.loopback = loopback_flag ? 1 : 0;
|
2019-05-28 17:59:39 -06:00
|
|
|
qstats.no_cn = no_cn_flag ? 1 : 0;
|
2019-03-01 13:38:49 -07:00
|
|
|
if (bpf_map_update_elem(map_fd, &key, &qstats, BPF_ANY)) {
|
|
|
|
printf("ERROR: Could not update map element\n");
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!outFlag)
|
|
|
|
type = BPF_CGROUP_INET_INGRESS;
|
|
|
|
if (bpf_prog_attach(bpfprog_fd, cg1, type, 0)) {
|
|
|
|
printf("ERROR: bpf_prog_attach fails!\n");
|
|
|
|
log_err("Attaching prog");
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (work_conserving_flag) {
|
|
|
|
struct timeval t0, t_last, t_new;
|
|
|
|
FILE *fin;
|
|
|
|
unsigned long long last_eth_tx_bytes, new_eth_tx_bytes;
|
|
|
|
signed long long last_cg_tx_bytes, new_cg_tx_bytes;
|
|
|
|
signed long long delta_time, delta_bytes, delta_rate;
|
|
|
|
int delta_ms;
|
|
|
|
#define DELTA_RATE_CHECK 10000 /* in us */
|
|
|
|
#define RATE_THRESHOLD 9500000000 /* 9.5 Gbps */
|
|
|
|
|
|
|
|
bpf_map_lookup_elem(map_fd, &key, &qstats);
|
|
|
|
if (gettimeofday(&t0, NULL) < 0)
|
|
|
|
do_error("gettimeofday failed", true);
|
|
|
|
t_last = t0;
|
|
|
|
fin = fopen("/sys/class/net/eth0/statistics/tx_bytes", "r");
|
|
|
|
if (fscanf(fin, "%llu", &last_eth_tx_bytes) != 1)
|
|
|
|
do_error("fscanf fails", false);
|
|
|
|
fclose(fin);
|
|
|
|
last_cg_tx_bytes = qstats.bytes_total;
|
|
|
|
while (true) {
|
|
|
|
usleep(DELTA_RATE_CHECK);
|
|
|
|
if (gettimeofday(&t_new, NULL) < 0)
|
|
|
|
do_error("gettimeofday failed", true);
|
|
|
|
delta_ms = (t_new.tv_sec - t0.tv_sec) * 1000 +
|
|
|
|
(t_new.tv_usec - t0.tv_usec)/1000;
|
|
|
|
if (delta_ms > dur * 1000)
|
|
|
|
break;
|
|
|
|
delta_time = (t_new.tv_sec - t_last.tv_sec) * 1000000 +
|
|
|
|
(t_new.tv_usec - t_last.tv_usec);
|
|
|
|
if (delta_time == 0)
|
|
|
|
continue;
|
|
|
|
t_last = t_new;
|
|
|
|
fin = fopen("/sys/class/net/eth0/statistics/tx_bytes",
|
|
|
|
"r");
|
|
|
|
if (fscanf(fin, "%llu", &new_eth_tx_bytes) != 1)
|
|
|
|
do_error("fscanf fails", false);
|
|
|
|
fclose(fin);
|
|
|
|
printf(" new_eth_tx_bytes:%llu\n",
|
|
|
|
new_eth_tx_bytes);
|
|
|
|
bpf_map_lookup_elem(map_fd, &key, &qstats);
|
|
|
|
new_cg_tx_bytes = qstats.bytes_total;
|
|
|
|
delta_bytes = new_eth_tx_bytes - last_eth_tx_bytes;
|
|
|
|
last_eth_tx_bytes = new_eth_tx_bytes;
|
|
|
|
delta_rate = (delta_bytes * 8000000) / delta_time;
|
|
|
|
printf("%5d - eth_rate:%.1fGbps cg_rate:%.3fGbps",
|
|
|
|
delta_ms, delta_rate/1000000000.0,
|
|
|
|
rate/1000.0);
|
|
|
|
if (delta_rate < RATE_THRESHOLD) {
|
|
|
|
/* can increase cgroup rate limit, but first
|
|
|
|
* check if we are using the current limit.
|
|
|
|
* Currently increasing by 6.25%, unknown
|
|
|
|
* if that is the optimal rate.
|
|
|
|
*/
|
|
|
|
int rate_diff100;
|
|
|
|
|
|
|
|
delta_bytes = new_cg_tx_bytes -
|
|
|
|
last_cg_tx_bytes;
|
|
|
|
last_cg_tx_bytes = new_cg_tx_bytes;
|
|
|
|
delta_rate = (delta_bytes * 8000000) /
|
|
|
|
delta_time;
|
|
|
|
printf(" rate:%.3fGbps",
|
|
|
|
delta_rate/1000000000.0);
|
|
|
|
rate_diff100 = (((long long)rate)*1000000 -
|
|
|
|
delta_rate) * 100 /
|
|
|
|
(((long long) rate) * 1000000);
|
|
|
|
printf(" rdiff:%d", rate_diff100);
|
|
|
|
if (rate_diff100 <= 3) {
|
|
|
|
rate += (rate >> 4);
|
|
|
|
if (rate > RATE_THRESHOLD / 1000000)
|
|
|
|
rate = RATE_THRESHOLD / 1000000;
|
|
|
|
qstats.rate = rate;
|
|
|
|
printf(" INC\n");
|
|
|
|
} else {
|
|
|
|
printf("\n");
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* Need to decrease cgroup rate limit.
|
|
|
|
* Currently decreasing by 12.5%, unknown
|
|
|
|
* if that is optimal
|
|
|
|
*/
|
|
|
|
printf(" DEC\n");
|
|
|
|
rate -= (rate >> 3);
|
|
|
|
if (rate < minRate)
|
|
|
|
rate = minRate;
|
|
|
|
qstats.rate = rate;
|
|
|
|
}
|
|
|
|
if (bpf_map_update_elem(map_fd, &key, &qstats, BPF_ANY))
|
|
|
|
do_error("update map element fails", false);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
sleep(dur);
|
|
|
|
}
|
|
|
|
// Get stats!
|
|
|
|
if (stats_flag && bpf_map_lookup_elem(map_fd, &key, &qstats)) {
|
|
|
|
char fname[100];
|
|
|
|
FILE *fout;
|
|
|
|
|
|
|
|
if (!outFlag)
|
|
|
|
sprintf(fname, "hbm.%d.in", cg_id);
|
|
|
|
else
|
|
|
|
sprintf(fname, "hbm.%d.out", cg_id);
|
|
|
|
fout = fopen(fname, "w");
|
|
|
|
fprintf(fout, "id:%d\n", cg_id);
|
|
|
|
fprintf(fout, "ERROR: Could not lookup queue_stats\n");
|
|
|
|
} else if (stats_flag && qstats.lastPacketTime >
|
|
|
|
qstats.firstPacketTime) {
|
|
|
|
long long delta_us = (qstats.lastPacketTime -
|
|
|
|
qstats.firstPacketTime)/1000;
|
|
|
|
unsigned int rate_mbps = ((qstats.bytes_total -
|
|
|
|
qstats.bytes_dropped) * 8 /
|
|
|
|
delta_us);
|
|
|
|
double percent_pkts, percent_bytes;
|
|
|
|
char fname[100];
|
|
|
|
FILE *fout;
|
2019-05-28 17:59:40 -06:00
|
|
|
int k;
|
|
|
|
static const char *returnValNames[] = {
|
|
|
|
"DROP_PKT",
|
|
|
|
"ALLOW_PKT",
|
|
|
|
"DROP_PKT_CWR",
|
|
|
|
"ALLOW_PKT_CWR"
|
|
|
|
};
|
|
|
|
#define RET_VAL_COUNT 4
|
2019-03-01 13:38:49 -07:00
|
|
|
|
|
|
|
// Future support of ingress
|
|
|
|
// if (!outFlag)
|
|
|
|
// sprintf(fname, "hbm.%d.in", cg_id);
|
|
|
|
// else
|
|
|
|
sprintf(fname, "hbm.%d.out", cg_id);
|
|
|
|
fout = fopen(fname, "w");
|
|
|
|
fprintf(fout, "id:%d\n", cg_id);
|
|
|
|
fprintf(fout, "rate_mbps:%d\n", rate_mbps);
|
|
|
|
fprintf(fout, "duration:%.1f secs\n",
|
|
|
|
(qstats.lastPacketTime - qstats.firstPacketTime) /
|
|
|
|
1000000000.0);
|
|
|
|
fprintf(fout, "packets:%d\n", (int)qstats.pkts_total);
|
|
|
|
fprintf(fout, "bytes_MB:%d\n", (int)(qstats.bytes_total /
|
|
|
|
1000000));
|
|
|
|
fprintf(fout, "pkts_dropped:%d\n", (int)qstats.pkts_dropped);
|
|
|
|
fprintf(fout, "bytes_dropped_MB:%d\n",
|
|
|
|
(int)(qstats.bytes_dropped /
|
|
|
|
1000000));
|
|
|
|
// Marked Pkts and Bytes
|
|
|
|
percent_pkts = (qstats.pkts_marked * 100.0) /
|
|
|
|
(qstats.pkts_total + 1);
|
|
|
|
percent_bytes = (qstats.bytes_marked * 100.0) /
|
|
|
|
(qstats.bytes_total + 1);
|
|
|
|
fprintf(fout, "pkts_marked_percent:%6.2f\n", percent_pkts);
|
|
|
|
fprintf(fout, "bytes_marked_percent:%6.2f\n", percent_bytes);
|
|
|
|
|
|
|
|
// Dropped Pkts and Bytes
|
|
|
|
percent_pkts = (qstats.pkts_dropped * 100.0) /
|
|
|
|
(qstats.pkts_total + 1);
|
|
|
|
percent_bytes = (qstats.bytes_dropped * 100.0) /
|
|
|
|
(qstats.bytes_total + 1);
|
|
|
|
fprintf(fout, "pkts_dropped_percent:%6.2f\n", percent_pkts);
|
|
|
|
fprintf(fout, "bytes_dropped_percent:%6.2f\n", percent_bytes);
|
2019-05-28 17:59:40 -06:00
|
|
|
|
|
|
|
// ECN CE markings
|
|
|
|
percent_pkts = (qstats.pkts_ecn_ce * 100.0) /
|
|
|
|
(qstats.pkts_total + 1);
|
|
|
|
fprintf(fout, "pkts_ecn_ce:%6.2f (%d)\n", percent_pkts,
|
|
|
|
(int)qstats.pkts_ecn_ce);
|
|
|
|
|
|
|
|
// Average cwnd
|
|
|
|
fprintf(fout, "avg cwnd:%d\n",
|
|
|
|
(int)(qstats.sum_cwnd / (qstats.sum_cwnd_cnt + 1)));
|
|
|
|
// Average rtt
|
|
|
|
fprintf(fout, "avg rtt:%d\n",
|
|
|
|
(int)(qstats.sum_rtt / (qstats.pkts_total + 1)));
|
|
|
|
// Average credit
|
bpf: Add support for fq's EDT to HBM
Adds support for fq's Earliest Departure Time to HBM (Host Bandwidth
Manager). Includes a new BPF program supporting EDT, and also updates
corresponding programs.
It will drop packets with an EDT of more than 500us in the future
unless the packet belongs to a flow with less than 2 packets in flight.
This is done so each flow has at least 2 packets in flight, so they
will not starve, and also to help prevent delayed ACK timeouts.
It will also work with ECN enabled traffic, where the packets will be
CE marked if their EDT is more than 50us in the future.
The table below shows some performance numbers. The flows are back to
back RPCS. One server sending to another, either 2 or 4 flows.
One flow is a 10KB RPC, the rest are 1MB RPCs. When there are more
than one flow of a given RPC size, the numbers represent averages.
The rate limit applies to all flows (they are in the same cgroup).
Tests ending with "-edt" ran with the new BPF program supporting EDT.
Tests ending with "-hbt" ran on top HBT qdisc with the specified rate
(i.e. no HBM). The other tests ran with the HBM BPF program included
in the HBM patch-set.
EDT has limited value when using DCTCP, but it helps in many cases when
using Cubic. It usually achieves larger link utilization and lower
99% latencies for the 1MB RPCs.
HBM ends up queueing a lot of packets with its default parameter values,
reducing the goodput of the 10KB RPCs and increasing their latency. Also,
the RTTs seen by the flows are quite large.
Aggr 10K 10K 10K 1MB 1MB 1MB
Limit rate drops RTT rate P90 P99 rate P90 P99
Test rate Flows Mbps % us Mbps us us Mbps ms ms
-------- ---- ----- ---- ----- --- ---- ---- ---- ---- ---- ----
cubic 1G 2 904 0.02 108 257 511 539 647 13.4 24.5
cubic-edt 1G 2 982 0.01 156 239 656 967 743 14.0 17.2
dctcp 1G 2 977 0.00 105 324 408 744 653 14.5 15.9
dctcp-edt 1G 2 981 0.01 142 321 417 811 660 15.7 17.0
cubic-htb 1G 2 919 0.00 1825 40 2822 4140 879 9.7 9.9
cubic 200M 2 155 0.30 220 81 532 655 74 283 450
cubic-edt 200M 2 188 0.02 222 87 1035 1095 101 84 85
dctcp 200M 2 188 0.03 111 77 912 939 111 76 325
dctcp-edt 200M 2 188 0.03 217 74 1416 1738 114 76 79
cubic-htb 200M 2 188 0.00 5015 8 14ms 15ms 180 48 50
cubic 1G 4 952 0.03 110 165 516 546 262 38 154
cubic-edt 1G 4 973 0.01 190 111 1034 1314 287 65 79
dctcp 1G 4 951 0.00 103 180 617 905 257 37 38
dctcp-edt 1G 4 967 0.00 163 151 732 1126 272 43 55
cubic-htb 1G 4 914 0.00 3249 13 7ms 8ms 300 29 34
cubic 5G 4 4236 0.00 134 305 490 624 1310 10 17
cubic-edt 5G 4 4865 0.00 156 306 425 759 1520 10 16
dctcp 5G 4 4936 0.00 128 485 221 409 1484 7 9
dctcp-edt 5G 4 4924 0.00 148 390 392 623 1508 11 26
v1 -> v2: Incorporated Andrii's suggestions
v2 -> v3: Incorporated Yonghong's suggestions
v3 -> v4: Removed credit update that is not needed
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-07-02 16:09:52 -06:00
|
|
|
if (edt_flag)
|
|
|
|
fprintf(fout, "avg credit_ms:%.03f\n",
|
|
|
|
(qstats.sum_credit /
|
|
|
|
(qstats.pkts_total + 1.0)) / 1000000.0);
|
|
|
|
else
|
|
|
|
fprintf(fout, "avg credit:%d\n",
|
|
|
|
(int)(qstats.sum_credit /
|
|
|
|
(1500 * ((int)qstats.pkts_total ) + 1)));
|
2019-05-28 17:59:40 -06:00
|
|
|
|
|
|
|
// Return values stats
|
|
|
|
for (k = 0; k < RET_VAL_COUNT; k++) {
|
|
|
|
percent_pkts = (qstats.returnValCount[k] * 100.0) /
|
|
|
|
(qstats.pkts_total + 1);
|
|
|
|
fprintf(fout, "%s:%6.2f (%d)\n", returnValNames[k],
|
|
|
|
percent_pkts, (int)qstats.returnValCount[k]);
|
|
|
|
}
|
2019-03-01 13:38:49 -07:00
|
|
|
fclose(fout);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (debugFlag)
|
|
|
|
read_trace_pipe2();
|
|
|
|
return rc;
|
|
|
|
err:
|
|
|
|
rc = 1;
|
|
|
|
|
|
|
|
if (cg1)
|
|
|
|
close(cg1);
|
|
|
|
cleanup_cgroup_environment();
|
|
|
|
|
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void Usage(void)
|
|
|
|
{
|
|
|
|
printf("This program loads a cgroup skb BPF program to enforce\n"
|
|
|
|
"cgroup output (egress) bandwidth limits.\n\n"
|
2019-05-28 17:59:39 -06:00
|
|
|
"USAGE: hbm [-o] [-d] [-l] [-n <id>] [--no_cn] [-r <rate>]\n"
|
|
|
|
" [-s] [-t <secs>] [-w] [-h] [prog]\n"
|
2019-03-01 13:38:49 -07:00
|
|
|
" Where:\n"
|
|
|
|
" -o indicates egress direction (default)\n"
|
|
|
|
" -d print BPF trace debug buffer\n"
|
bpf: Add support for fq's EDT to HBM
Adds support for fq's Earliest Departure Time to HBM (Host Bandwidth
Manager). Includes a new BPF program supporting EDT, and also updates
corresponding programs.
It will drop packets with an EDT of more than 500us in the future
unless the packet belongs to a flow with less than 2 packets in flight.
This is done so each flow has at least 2 packets in flight, so they
will not starve, and also to help prevent delayed ACK timeouts.
It will also work with ECN enabled traffic, where the packets will be
CE marked if their EDT is more than 50us in the future.
The table below shows some performance numbers. The flows are back to
back RPCS. One server sending to another, either 2 or 4 flows.
One flow is a 10KB RPC, the rest are 1MB RPCs. When there are more
than one flow of a given RPC size, the numbers represent averages.
The rate limit applies to all flows (they are in the same cgroup).
Tests ending with "-edt" ran with the new BPF program supporting EDT.
Tests ending with "-hbt" ran on top HBT qdisc with the specified rate
(i.e. no HBM). The other tests ran with the HBM BPF program included
in the HBM patch-set.
EDT has limited value when using DCTCP, but it helps in many cases when
using Cubic. It usually achieves larger link utilization and lower
99% latencies for the 1MB RPCs.
HBM ends up queueing a lot of packets with its default parameter values,
reducing the goodput of the 10KB RPCs and increasing their latency. Also,
the RTTs seen by the flows are quite large.
Aggr 10K 10K 10K 1MB 1MB 1MB
Limit rate drops RTT rate P90 P99 rate P90 P99
Test rate Flows Mbps % us Mbps us us Mbps ms ms
-------- ---- ----- ---- ----- --- ---- ---- ---- ---- ---- ----
cubic 1G 2 904 0.02 108 257 511 539 647 13.4 24.5
cubic-edt 1G 2 982 0.01 156 239 656 967 743 14.0 17.2
dctcp 1G 2 977 0.00 105 324 408 744 653 14.5 15.9
dctcp-edt 1G 2 981 0.01 142 321 417 811 660 15.7 17.0
cubic-htb 1G 2 919 0.00 1825 40 2822 4140 879 9.7 9.9
cubic 200M 2 155 0.30 220 81 532 655 74 283 450
cubic-edt 200M 2 188 0.02 222 87 1035 1095 101 84 85
dctcp 200M 2 188 0.03 111 77 912 939 111 76 325
dctcp-edt 200M 2 188 0.03 217 74 1416 1738 114 76 79
cubic-htb 200M 2 188 0.00 5015 8 14ms 15ms 180 48 50
cubic 1G 4 952 0.03 110 165 516 546 262 38 154
cubic-edt 1G 4 973 0.01 190 111 1034 1314 287 65 79
dctcp 1G 4 951 0.00 103 180 617 905 257 37 38
dctcp-edt 1G 4 967 0.00 163 151 732 1126 272 43 55
cubic-htb 1G 4 914 0.00 3249 13 7ms 8ms 300 29 34
cubic 5G 4 4236 0.00 134 305 490 624 1310 10 17
cubic-edt 5G 4 4865 0.00 156 306 425 759 1520 10 16
dctcp 5G 4 4936 0.00 128 485 221 409 1484 7 9
dctcp-edt 5G 4 4924 0.00 148 390 392 623 1508 11 26
v1 -> v2: Incorporated Andrii's suggestions
v2 -> v3: Incorporated Yonghong's suggestions
v3 -> v4: Removed credit update that is not needed
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-07-02 16:09:52 -06:00
|
|
|
" --edt use fq's Earliest Departure Time\n"
|
2019-03-01 13:38:49 -07:00
|
|
|
" -l also limit flows using loopback\n"
|
|
|
|
" -n <#> to create cgroup \"/hbm#\" and attach prog\n"
|
|
|
|
" Default is /hbm1\n"
|
2019-06-03 07:36:53 -06:00
|
|
|
" --no_cn disable CN notifications\n"
|
2019-03-01 13:38:49 -07:00
|
|
|
" -r <rate> Rate in Mbps\n"
|
|
|
|
" -s Update HBM stats\n"
|
2019-03-05 10:31:13 -07:00
|
|
|
" -t <time> Exit after specified seconds (default is 0)\n"
|
2019-03-01 13:38:49 -07:00
|
|
|
" -w Work conserving flag. cgroup can increase\n"
|
|
|
|
" bandwidth beyond the rate limit specified\n"
|
|
|
|
" while there is available bandwidth. Current\n"
|
|
|
|
" implementation assumes there is only eth0\n"
|
|
|
|
" but can be extended to support multiple NICs\n"
|
|
|
|
" -h print this info\n"
|
|
|
|
" prog BPF program file name. Name defaults to\n"
|
|
|
|
" hbm_out_kern.o\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
int main(int argc, char **argv)
|
|
|
|
{
|
|
|
|
char *prog = "hbm_out_kern.o";
|
|
|
|
int k;
|
|
|
|
int cg_id = 1;
|
|
|
|
char *optstring = "iodln:r:st:wh";
|
2019-05-28 17:59:39 -06:00
|
|
|
struct option loptions[] = {
|
|
|
|
{"no_cn", 0, NULL, 1},
|
bpf: Add support for fq's EDT to HBM
Adds support for fq's Earliest Departure Time to HBM (Host Bandwidth
Manager). Includes a new BPF program supporting EDT, and also updates
corresponding programs.
It will drop packets with an EDT of more than 500us in the future
unless the packet belongs to a flow with less than 2 packets in flight.
This is done so each flow has at least 2 packets in flight, so they
will not starve, and also to help prevent delayed ACK timeouts.
It will also work with ECN enabled traffic, where the packets will be
CE marked if their EDT is more than 50us in the future.
The table below shows some performance numbers. The flows are back to
back RPCS. One server sending to another, either 2 or 4 flows.
One flow is a 10KB RPC, the rest are 1MB RPCs. When there are more
than one flow of a given RPC size, the numbers represent averages.
The rate limit applies to all flows (they are in the same cgroup).
Tests ending with "-edt" ran with the new BPF program supporting EDT.
Tests ending with "-hbt" ran on top HBT qdisc with the specified rate
(i.e. no HBM). The other tests ran with the HBM BPF program included
in the HBM patch-set.
EDT has limited value when using DCTCP, but it helps in many cases when
using Cubic. It usually achieves larger link utilization and lower
99% latencies for the 1MB RPCs.
HBM ends up queueing a lot of packets with its default parameter values,
reducing the goodput of the 10KB RPCs and increasing their latency. Also,
the RTTs seen by the flows are quite large.
Aggr 10K 10K 10K 1MB 1MB 1MB
Limit rate drops RTT rate P90 P99 rate P90 P99
Test rate Flows Mbps % us Mbps us us Mbps ms ms
-------- ---- ----- ---- ----- --- ---- ---- ---- ---- ---- ----
cubic 1G 2 904 0.02 108 257 511 539 647 13.4 24.5
cubic-edt 1G 2 982 0.01 156 239 656 967 743 14.0 17.2
dctcp 1G 2 977 0.00 105 324 408 744 653 14.5 15.9
dctcp-edt 1G 2 981 0.01 142 321 417 811 660 15.7 17.0
cubic-htb 1G 2 919 0.00 1825 40 2822 4140 879 9.7 9.9
cubic 200M 2 155 0.30 220 81 532 655 74 283 450
cubic-edt 200M 2 188 0.02 222 87 1035 1095 101 84 85
dctcp 200M 2 188 0.03 111 77 912 939 111 76 325
dctcp-edt 200M 2 188 0.03 217 74 1416 1738 114 76 79
cubic-htb 200M 2 188 0.00 5015 8 14ms 15ms 180 48 50
cubic 1G 4 952 0.03 110 165 516 546 262 38 154
cubic-edt 1G 4 973 0.01 190 111 1034 1314 287 65 79
dctcp 1G 4 951 0.00 103 180 617 905 257 37 38
dctcp-edt 1G 4 967 0.00 163 151 732 1126 272 43 55
cubic-htb 1G 4 914 0.00 3249 13 7ms 8ms 300 29 34
cubic 5G 4 4236 0.00 134 305 490 624 1310 10 17
cubic-edt 5G 4 4865 0.00 156 306 425 759 1520 10 16
dctcp 5G 4 4936 0.00 128 485 221 409 1484 7 9
dctcp-edt 5G 4 4924 0.00 148 390 392 623 1508 11 26
v1 -> v2: Incorporated Andrii's suggestions
v2 -> v3: Incorporated Yonghong's suggestions
v3 -> v4: Removed credit update that is not needed
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-07-02 16:09:52 -06:00
|
|
|
{"edt", 0, NULL, 2},
|
2019-05-28 17:59:39 -06:00
|
|
|
{NULL, 0, NULL, 0}
|
|
|
|
};
|
2019-03-01 13:38:49 -07:00
|
|
|
|
2019-05-28 17:59:39 -06:00
|
|
|
while ((k = getopt_long(argc, argv, optstring, loptions, NULL)) != -1) {
|
2019-03-01 13:38:49 -07:00
|
|
|
switch (k) {
|
2019-05-28 17:59:39 -06:00
|
|
|
case 1:
|
|
|
|
no_cn_flag = true;
|
|
|
|
break;
|
bpf: Add support for fq's EDT to HBM
Adds support for fq's Earliest Departure Time to HBM (Host Bandwidth
Manager). Includes a new BPF program supporting EDT, and also updates
corresponding programs.
It will drop packets with an EDT of more than 500us in the future
unless the packet belongs to a flow with less than 2 packets in flight.
This is done so each flow has at least 2 packets in flight, so they
will not starve, and also to help prevent delayed ACK timeouts.
It will also work with ECN enabled traffic, where the packets will be
CE marked if their EDT is more than 50us in the future.
The table below shows some performance numbers. The flows are back to
back RPCS. One server sending to another, either 2 or 4 flows.
One flow is a 10KB RPC, the rest are 1MB RPCs. When there are more
than one flow of a given RPC size, the numbers represent averages.
The rate limit applies to all flows (they are in the same cgroup).
Tests ending with "-edt" ran with the new BPF program supporting EDT.
Tests ending with "-hbt" ran on top HBT qdisc with the specified rate
(i.e. no HBM). The other tests ran with the HBM BPF program included
in the HBM patch-set.
EDT has limited value when using DCTCP, but it helps in many cases when
using Cubic. It usually achieves larger link utilization and lower
99% latencies for the 1MB RPCs.
HBM ends up queueing a lot of packets with its default parameter values,
reducing the goodput of the 10KB RPCs and increasing their latency. Also,
the RTTs seen by the flows are quite large.
Aggr 10K 10K 10K 1MB 1MB 1MB
Limit rate drops RTT rate P90 P99 rate P90 P99
Test rate Flows Mbps % us Mbps us us Mbps ms ms
-------- ---- ----- ---- ----- --- ---- ---- ---- ---- ---- ----
cubic 1G 2 904 0.02 108 257 511 539 647 13.4 24.5
cubic-edt 1G 2 982 0.01 156 239 656 967 743 14.0 17.2
dctcp 1G 2 977 0.00 105 324 408 744 653 14.5 15.9
dctcp-edt 1G 2 981 0.01 142 321 417 811 660 15.7 17.0
cubic-htb 1G 2 919 0.00 1825 40 2822 4140 879 9.7 9.9
cubic 200M 2 155 0.30 220 81 532 655 74 283 450
cubic-edt 200M 2 188 0.02 222 87 1035 1095 101 84 85
dctcp 200M 2 188 0.03 111 77 912 939 111 76 325
dctcp-edt 200M 2 188 0.03 217 74 1416 1738 114 76 79
cubic-htb 200M 2 188 0.00 5015 8 14ms 15ms 180 48 50
cubic 1G 4 952 0.03 110 165 516 546 262 38 154
cubic-edt 1G 4 973 0.01 190 111 1034 1314 287 65 79
dctcp 1G 4 951 0.00 103 180 617 905 257 37 38
dctcp-edt 1G 4 967 0.00 163 151 732 1126 272 43 55
cubic-htb 1G 4 914 0.00 3249 13 7ms 8ms 300 29 34
cubic 5G 4 4236 0.00 134 305 490 624 1310 10 17
cubic-edt 5G 4 4865 0.00 156 306 425 759 1520 10 16
dctcp 5G 4 4936 0.00 128 485 221 409 1484 7 9
dctcp-edt 5G 4 4924 0.00 148 390 392 623 1508 11 26
v1 -> v2: Incorporated Andrii's suggestions
v2 -> v3: Incorporated Yonghong's suggestions
v3 -> v4: Removed credit update that is not needed
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-07-02 16:09:52 -06:00
|
|
|
case 2:
|
|
|
|
prog = "hbm_edt_kern.o";
|
|
|
|
edt_flag = true;
|
|
|
|
break;
|
2019-03-01 13:38:49 -07:00
|
|
|
case'o':
|
|
|
|
break;
|
|
|
|
case 'd':
|
|
|
|
debugFlag = true;
|
|
|
|
break;
|
|
|
|
case 'l':
|
|
|
|
loopback_flag = true;
|
|
|
|
break;
|
|
|
|
case 'n':
|
|
|
|
cg_id = atoi(optarg);
|
|
|
|
break;
|
|
|
|
case 'r':
|
|
|
|
minRate = atoi(optarg) * 1.024;
|
|
|
|
rate = minRate;
|
|
|
|
break;
|
|
|
|
case 's':
|
|
|
|
stats_flag = true;
|
|
|
|
break;
|
|
|
|
case 't':
|
|
|
|
dur = atoi(optarg);
|
|
|
|
break;
|
|
|
|
case 'w':
|
|
|
|
work_conserving_flag = true;
|
|
|
|
break;
|
|
|
|
case '?':
|
|
|
|
if (optopt == 'n' || optopt == 'r' || optopt == 't')
|
|
|
|
fprintf(stderr,
|
|
|
|
"Option -%c requires an argument.\n\n",
|
|
|
|
optopt);
|
|
|
|
case 'h':
|
|
|
|
// fallthrough
|
|
|
|
default:
|
|
|
|
Usage();
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (optind < argc)
|
|
|
|
prog = argv[optind];
|
|
|
|
printf("HBM prog: %s\n", prog != NULL ? prog : "NULL");
|
|
|
|
|
|
|
|
return run_bpf_prog(prog, cg_id);
|
|
|
|
}
|