add cpu usage report, add multi core support. Add optional undefined …

.gitignore

@@ -3,3 +3,5 @@
 build*
 *.cache
 compile_commands.json
+test_client.sh
+test_server.sh

CMakeLists.txt

@@ -3,20 +3,45 @@ project(qperf VERSION 1.0.0 LANGUAGES C)
 
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
+# Set linker flags to use lld
+if(DEFINED SANITIZE)
+    set(CMAKE_EXE_LINKER_FLAGS "-fuse-ld=lld")
+    set(CMAKE_MODULE_LINKER_FLAGS "-fuse-ld=lld")
+    set(CMAKE_SHARED_LINKER_FLAGS "-fuse-ld=lld")
+
+    # Set the sanitize options
+    set(SANITIZE_FLAGS
+     -fsanitize=address
+     -fsanitize=undefined
+     -fsanitize=signed-integer-overflow
+     -fsanitize=shift
+     -fsanitize=integer-divide-by-zero
+     -fsanitize=array-bounds
+     -fsanitize=implicit-conversion
+     -fsanitize=nullability
+     -fsanitize=integer
+     -fsanitize-address-use-after-return=always
+     -fsanitize-address-use-after-scope
+    )
+    set(BETTER_STACKTRACE -g -O1 -fno-omit-frame-pointer -fno-optimize-sibling-calls)
+endif()
+
 add_subdirectory(extern)
 
 add_executable(qperf main.c
     client.h client.c
     client_stream.h client_stream.c
     server.h server.c
     server_stream.h server_stream.c
-    common.h common.c)
+    common.h common.c
+    cpu-usage.h cpu-usage.c)
 
-target_link_libraries(qperf PRIVATE quicly ev picotls)
+target_link_libraries(qperf PRIVATE quicly ev picotls ${SANITIZE_FLAGS} ${BETTER_STACKTRACE})
 target_compile_definitions(qperf PRIVATE QPERF_VERSION="${PROJECT_VERSION}")
 target_compile_options(qperf PRIVATE
     -Werror=implicit-function-declaration
     -Werror=incompatible-pointer-types
     -Werror=shift-count-overflow
+    ${SANITIZE_FLAGS} ${BETTER_STACKTRACE}
 )
 

README.md

@@ -7,56 +7,67 @@ Uses https://github.com/h2o/quicly
 Usage: ./qperf [options]
 
 Options:
-  -c target             run as client and connect to target server
-  --cc [reno,cubic]     congestion control algorithm to use (default reno)
-  -e                    measure time for connection establishment and first byte only
-  -g                    enable UDP generic segmentation offload
-  --iw initial-window   initial window to use (default 10)
-  -l log-file           file to log tls secrets
-  -p                    port to listen on/connect to (default 18080)
-  -s                    run as server
-  -t time (s)           run for X seconds (default 10s)
-  -h                    print this help
+  -c target            run as client and connect to target server
+  --cc [reno,cubic]    congestion control algorithm to use (default reno)
+  -e                   measure time for connection establishment and first byte only
+  -g                   enable UDP generic segmentation offload
+  --iw initial-window  initial window to use (default 10)
+  -l log-file          file to log tls secrets
+  -p                   port to listen on/connect to (default 18080)
+  -s  address          listen as server on address
+  -t time (s)          run for X seconds (default 10s)
+  -i num_instances     number of instances to start (default nproc)
+  -h                   print this help
 ```
 
-server
+
+#### client
 ```
-./qperf -s
-starting server with pid 5624 on port 18080
-got new connection
-request received, sending data
-connection 0 second 0 send window: 1112923 packets sent: 364792 packets lost: 373
-connection 0 second 1 send window: 1238055 packets sent: 377515 packets lost: 123
-connection 0 second 2 send window: 583352 packets sent: 355482 packets lost: 862
-connection 0 second 3 send window: 275563 packets sent: 367538 packets lost: 607
-connection 0 second 4 send window: 1100261 packets sent: 366005 packets lost: 20
-connection 0 second 5 send window: 633010 packets sent: 356021 packets lost: 857
-connection 0 second 6 send window: 1266610 packets sent: 367866 packets lost: 0
-connection 0 second 7 send window: 1668530 packets sent: 360649 packets lost: 0
-connection 0 second 8 send window: 1994930 packets sent: 364087 packets lost: 0
-connection 0 second 9 send window: 1779683 packets sent: 374804 packets lost: 80
-connection 0 total packets sent: 3654759 total packets lost: 2922
+./qperf -c 2a01:4f8:1c1b:fd7a::1 -p 18000 -n 1
+starting client with host 2a01:4f8:1c1b:fd7a::1, port 18000, runtime 10s, cc reno, iw 10
+connection establishment time: 16ms
+time to first byte: 16ms
+second 0: 1.158 gbit/s, cpu 0: 21.65%
+second 1: 1.161 gbit/s, cpu 2: 50.00%
+second 2: 1.205 gbit/s, cpu 0: 32.29%
+second 3: 1.162 gbit/s, cpu 2: 52.87%
+second 4: 1.258 gbit/s, cpu 1: 34.04%
+second 5: 1.387 gbit/s, cpu 1: 50.00%
+second 6: 1.393 gbit/s, cpu 2: 15.96%
+second 7: 1.366 gbit/s, cpu 0: 17.35%
+second 8: 1.195 gbit/s, cpu 0: 46.67%
+second 9: 1.226 gbit/s, cpu 1: 15.62%
+connection closed
 ```
-*Note*: The server looks for a TLS certificate and key in the current working dir named "server.crt" and "server.key" respectively([See TLS](#TLS)). You can use a self signed certificate; the client doesn't validate it.
 
 
-client
+#### server
+
 ```
-./qperf -c 127.0.0.1
-running client with host=127.0.0.1 and runtime=10s
-connection establishment time: 6ms
-time to first byte: 7ms
-second 0: 3.144 gbit/s (422030372 bytes received)
-second 1: 3.444 gbit/s (462189378 bytes received)
-second 2: 3.184 gbit/s (427337822 bytes received)
-second 3: 3.333 gbit/s (447304096 bytes received)
-second 4: 2.996 gbit/s (402100242 bytes received)
-second 5: 3.274 gbit/s (439462608 bytes received)
-second 6: 3.083 gbit/s (413746021 bytes received)
-second 7: 3.336 gbit/s (447686682 bytes received)
-second 8: 3.034 gbit/s (407235597 bytes received)
-second 9: 3.02 gbit/s (405314061 bytes received)
+./qperf -s :: -p 18000
+starting server with pid 15291,address ::, port 18000, cc reno, iw 10
+starting server with pid 15293,address ::, port 18002, cc reno, iw 10
+starting server with pid 15292,address ::, port 18001, cc reno, iw 10
+starting server with pid 15294,address ::, port 18003, cc reno, iw 10
+got new connection
+request received, sending data
+connection 0 second 0 send window: 622682 packets sent: 133403 packets lost: 2289, cpu 3: 1.99%
+connection 0 second 1 send window: 716581 packets sent: 126095 packets lost: 23, cpu 3: 98.99%
+connection 0 second 2 send window: 652943 packets sent: 131597 packets lost: 37, cpu 3: 97.98%
+connection 0 second 3 send window: 563577 packets sent: 138904 packets lost: 35, cpu 3: 96.97%
+connection 0 second 4 send window: 637191 packets sent: 127947 packets lost: 43, cpu 3: 100.00%
+connection 0 second 5 send window: 727889 packets sent: 119218 packets lost: 13, cpu 3: 99.00%
+connection 0 second 6 send window: 577401 packets sent: 111255 packets lost: 62, cpu 2: 53.33%
+connection 0 second 7 send window: 807801 packets sent: 97623 packets lost: 0, cpu 2: 100.00%
+connection 0 second 8 send window: 976761 packets sent: 92145 packets lost: 0, cpu 3: 53.54%
+connection 0 second 9 send window: 890324 packets sent: 104484 packets lost: 11, cpu 3: 100.00%
+transport close:code=0x0;frame=0;reason=
+connection 0 second 10 send window: 891604 packets sent: 0 packets lost: 0, cpu 3: -nan%
+connection 0 total packets sent: 1182671 total packets lost: 2513
+connection closed
 ```
+*Note*: The server looks for a TLS certificate and key in the current working dir named "server.crt" and "server.key" respectively([See TLS](#TLS)). You can use a self signed certificate; the client doesn't validate it.
+
 
 # how to build
 ## 1. Install required dependencies 

client.c

@@ -1,6 +1,7 @@
 #include "client.h"
 #include "client_stream.h"
 #include "common.h"
+#include "cpu-usage.h"
 
 #include <ev.h>
 #include <stdio.h>
@@ -13,6 +14,7 @@
 #include <float.h>
 #include <quicly/streambuf.h>
 
+
 #include <picotls/../../t/util.h>
 
 static int client_socket = -1;
@@ -125,6 +127,7 @@ int run_client(const char *port, bool gso, const char *logfile, const char *cc,
 {
     setup_session_cache(get_tlsctx());
     quicly_amend_ptls_context(get_tlsctx());
+    getcpuusage();
 
     client_ctx = quicly_spec_context;
     client_ctx.tls = get_tlsctx();

client_stream.c

@@ -1,9 +1,17 @@
+
+#define _GNU_SOURCE
 #include "client_stream.h"
 #include "client.h"
 #include "common.h"
 #include <ev.h>
 #include <stdbool.h>
 #include <quicly/streambuf.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sched.h>
+#include "cpu-usage.h"
+
+
 
 static int current_second = 0;
 static uint64_t bytes_received = 0;
@@ -28,8 +36,15 @@ static void report_cb(EV_P_ ev_timer *w, int revents)
 {
     char size_str[100];
     format_size(size_str, bytes_received);
+    double *percent = getcpuusage();
+
+    int cpu_core = sched_getcpu();
+    if (cpu_core == -1) {
+        perror("sched_getcpu");
+        exit(1);
+    }
 
-    printf("second %i: %s (%lu bytes received)\n", current_second, size_str, bytes_received);
+    printf("second %i: %s, cpu %d: %.2f%%\n", current_second, size_str, cpu_core, percent[cpu_core]);
     fflush(stdout);
     ++current_second;
     bytes_received = 0;

common.c

@@ -1,5 +1,4 @@
 #include "common.h"
-
 #include <sys/socket.h>
 #include <netinet/udp.h>
 #include <netdb.h>
@@ -22,8 +21,6 @@ struct addrinfo *get_address(const char *host, const char *port)
     struct addrinfo hints;
     struct addrinfo *result;
 
-    printf("resolving %s:%s\n", host, port);
-
     memset(&hints, 0, sizeof(struct addrinfo));
 
     hints.ai_family = AF_UNSPEC; // Let getaddrinfo decide if it's a hostname.

common.h

@@ -6,14 +6,15 @@
 #include <unistd.h>
 #include <sys/syscall.h>
 
+#define BREAKPOINT printf("%s() at %s:%d\n", __func__, __FILE__, __LINE__); __asm__("int3; nop" ::: "memory");
+
 ptls_context_t *get_tlsctx();
 
 struct addrinfo *get_address(const char *host, const char *port);
 void enable_gso();
 bool send_pending(quicly_context_t *ctx, int fd, quicly_conn_t *conn);
 void print_escaped(const char *src, size_t len);
 
-
 static inline int64_t min_int64(int64_t a, int64_t b)
 {
     if(a < b) {
@@ -44,7 +45,7 @@ static inline int64_t clamp_int64(int64_t val, int64_t min, int64_t max)
 
 static inline uint64_t get_current_pid()
 {
-    uint64_t pid;
+    uint64_t pid;   
 
     #ifdef __APPLE__
         pthread_threadid_np(NULL, &pid);
@@ -53,4 +54,5 @@ static inline uint64_t get_current_pid()
     #endif
 
     return pid;
-}
+}
+