Skip to contentSkip to conversation

vnc: throttle screen updates.

PATCH In review Gerd Hoffmann 2 participants 4 messages Mar 18, 2009 – Mar 20, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 3/3 vnc: throttle screen updates.

Hi,

This patch makes the vnc server code skip screen refreshes in case there is data in the output buffer. This reduces the refresh rate to throttle the bandwidth needed in case the network link is saturated.

cheers,
Gerd

Quoted 1 line

From e1a20de776afe0d108fa94b2170f6d5690f9a24c Mon Sep 17 00:00:00 2001

From: Gerd Hoffmann <[email protected]>
Date: Mon, 16 Mar 2009 21:37:48 +0100
Subject: [PATCH 3/3] vnc: throttle screen updates.

This patch makes the vnc server code skip screen refreshes in case there is data in the output buffer. This reduces the refresh rate to throttle the bandwidth needed in case the network link is saturated.

---
 vnc.c |   11 ++++++++++-
 vnc.h |    1 +
 2 files changed, 11 insertions(+), 1 deletions(-)
vnc.c +10 -1
index 0daaeac..8c60547 100644
@@ -649,6 +649,7 @@ static void send_framebuffer_update(VncState *vs, int x, int y, int w, int h)
 
 static void vnc_copy(VncState *vs, int src_x, int src_y, int dst_x, int dst_y, int w, int h)
 {
+    vs->force_update = 1;
     vnc_update_client(vs);
 
     vnc_write_u8(vs, 0);  /* msg id */
@@ -704,6 +705,12 @@ static void vnc_update_client(void *opaque)
 
         vga_hw_update();
 
+        if (vs->output.offset && !vs->audio_cap && !vs->force_update) {
+            /* kernel send buffers are full -> drop frames to throttle */
+            qemu_mod_timer(vs->timer, qemu_get_clock(rt_clock) + VNC_REFRESH_INTERVAL);
+            return;
+        }
+
         /*
          * Walk through the guest dirty map.
          * Check and copy modified bits from guest to server surface.
@@ -737,7 +744,7 @@ static void vnc_update_client(void *opaque)
             server_row += ds_get_linesize(vs->ds);
         }
 
-        if (!has_dirty && !vs->audio_cap) {
+        if (!has_dirty && !vs->audio_cap && !vs->force_update) {
             qemu_mod_timer(vs->timer, qemu_get_clock(rt_clock) + VNC_REFRESH_INTERVAL);
             return;
         }
@@ -781,6 +788,7 @@ static void vnc_update_client(void *opaque)
         vs->output.buffer[saved_offset] = (n_rectangles >> 8) & 0xFF;
         vs->output.buffer[saved_offset + 1] = n_rectangles & 0xFF;
         vnc_flush(vs);
+        vs->force_update = 0;
 
     }
 
@@ -1398,6 +1406,7 @@ static void framebuffer_update_request(VncState *vs, int incremental,
 
     int i;
     vs->need_update = 1;
+    vs->force_update = 1;
     if (!incremental) {
         for (i = 0; i < h; i++) {
             vnc_set_bits(vs->guest.dirty[y_position + i],
vnc.h +1 -0
index c00fb00..0a2c66c 100644
@@ -121,6 +121,7 @@ struct VncState
 
     VncDisplay *vd;
     int need_update;
+    int force_update;
     uint32_t features;
     int absolute;
     int last_x;
Signature · 1 line
1.6.1.3
#1 lore raw
Quoted 6 lines · Gerd Hoffmann

Hi,

This patch makes the vnc server code skip screen refreshes in case there is data in the output buffer. This reduces the refresh rate to throttle the bandwidth needed in case the network link is saturated.

I like the idea behind this patch, but I would make force_update an unsigned integer instead of a boolean, incrementing it every time we get a framebuffer_update_request and decrementing it every time we send a reply.

Secondly we could even move vga_hw_update after:

if (vs->output.offset && !vs->audio_cap && !vs->force_update)

#2 lore raw
Quoted 4 lines · Stefano Stabellini

I like the idea behind this patch, but I would make force_update an unsigned integer instead of a boolean, incrementing it every time we get a framebuffer_update_request and decrementing it every time we send a reply.

Point being?
Note that we can satisfy multiple update requests in one reply. Also note that vnc_copy() sets force_update too.

Quoted 3 lines

Secondly we could even move vga_hw_update after:

if (vs->output.offset && !vs->audio_cap && !vs->force_update)

Agreed.

cheers,
Gerd

#3 lore raw
Quoted 3 lines · Gerd Hoffmann

Point being?
Note that we can satisfy multiple update requests in one reply. Also note that vnc_copy() sets force_update too.

right, not so useful then. I drop the suggestion.