[KERNEL] Slow down with RTT measurements after 30 seconds

Right after connecting, all servers will be polled every 4 seconds for 30 seconds,
so we get 7 data points per alt-server. If no better server is found during
this time, further RTT measurements will be done every 22 seconds, to put
less load on the network in the long run.

3 files changed, 26 insertions, 8 deletions

diff --git a/src/config.h b/src/config.h
index 128edb7..7df49a0 100644
--- a/src/config.h
+++ b/src/config.h
@@ -71,10 +71,12 @@
 // This must be a power of two:
 #define RTT_BLOCK_SIZE 4096
 
+#define STARTUP_MODE_DURATION 30
 // Interval of several repeating tasks (in seconds)
-#define TIMER_INTERVAL_PROBE_NORMAL 10
+#define TIMER_INTERVAL_PROBE_STARTUP 4
+#define TIMER_INTERVAL_PROBE_NORMAL 22
 #define TIMER_INTERVAL_PROBE_PANIC 2
-#define TIMER_INTERVAL_KEEPALIVE_PACKET 5
+#define TIMER_INTERVAL_KEEPALIVE_PACKET 6
 
 // Expect a keepalive response every X seconds
 #define SOCKET_KEEPALIVE_TIMEOUT 7
diff --git a/src/kernel/dnbd3.h b/src/kernel/dnbd3.h
index 915c8b9..aa9ea86 100644
--- a/src/kernel/dnbd3.h
+++ b/src/kernel/dnbd3.h
@@ -61,7 +61,8 @@ typedef struct
 	int new_servers_num;	// number of new alt servers that are waiting to be copied to above array
 	dnbd3_server_entry_t new_servers[NUMBER_SERVERS]; // pending new alt servers
 	uint8_t discover, panic, disconnecting, is_server, update_available, panic_count;
-	uint16_t rid, heartbeat_count;
+	uint16_t rid;
+	uint32_t heartbeat_count;
 	uint64_t reported_size;
 	// server switch
 	struct socket *better_sock;
diff --git a/src/kernel/net.c b/src/kernel/net.c
index b3e08df..fc237de 100644
--- a/src/kernel/net.c
+++ b/src/kernel/net.c
@@ -394,7 +394,8 @@ void dnbd3_net_heartbeat(unsigned long arg)
 				debug_dev("ERROR: Couldn't create keepalive request.");
 			}
 		}
-		if (timeout_seconds(TIMER_INTERVAL_PROBE_NORMAL))
+		if ((dev->heartbeat_count > STARTUP_MODE_DURATION && timeout_seconds(TIMER_INTERVAL_PROBE_NORMAL))
+				|| (dev->heartbeat_count <= STARTUP_MODE_DURATION && timeout_seconds(TIMER_INTERVAL_PROBE_STARTUP)))
 		{
 			// Normal discovery
 			dev->discover = 1;
@@ -436,9 +437,9 @@ int dnbd3_net_discover(void *data)
 	struct timeval start, end;
 	unsigned long rtt, best_rtt = 0;
 	unsigned long irqflags;
-	int i, best_server, current_server;
+	int i, istart, isize, best_server, current_server;
 	int turn = 0;
-	int ready = 0, do_change;
+	int ready = 0, do_change, last_alt_count = 0;
 	int mlen;
 
 	struct request *last_request = (struct request *)123, *cur_request = (struct request *)456;
@@ -524,12 +525,26 @@ int dnbd3_net_discover(void *data)
 		current_server = best_server = -1;
 		best_rtt = 0xFFFFFFFul;
 
-		for (i = 0; i < NUMBER_SERVERS; ++i)
+		if (dev->heartbeat_count < STARTUP_MODE_DURATION || last_alt_count == 0 || dev->panic)
+		{
+			istart = 0;
+			isize = NUMBER_SERVERS;
+		}
+		else
+		{
+			istart = jiffies % MAX(last_alt_count - 2, 1);
+			isize = 3;
+		}
+
+		for (i = istart; i < NUMBER_SERVERS; ++i)
 		{
 			if (dev->alt_servers[i].host.type == 0) // Empty slot
 				continue;
+			last_alt_count = i;
 			if (!dev->panic && dev->alt_servers[i].failures > 50 && (jiffies & 7) != 0) // If not in panic mode, skip server if it failed too many times
 				continue;
+			if (isize-- <= 0)
+				break;
 
 			// Initialize socket and connect
 			if (sock_create_kern(dev->alt_servers[i].host.type, SOCK_STREAM, IPPROTO_TCP, &sock) < 0)
@@ -783,7 +798,7 @@ int dnbd3_net_discover(void *data)
 			best_sock = NULL;
 		}
 
-		if (!ready || (jiffies & 3) != 0)
+		if (!ready || (jiffies & 7) != 0)
 			turn = (turn + 1) % 4;
 		if (turn == 3)
 			ready = 1;