update performance to 4k, when mr_sleep to 2000ms. 2.0.52

@@ -738,6 +738,7 @@ The publish benchmark by [st-load](https://github.com/winlinvip/st-load):
 * 2014-12-04, SRS 2.0.49,  1.4k(1400) publishers, 68%CPU, 144MB.
 * 2014-12-04, SRS 2.0.49,  2.5k(2500) publishers, 95%CPU, 404MB. [commit](https://github.com/winlinvip/simple-rtmp-server/commit/29324fab469e0f7cef9ad04ffdbce832ac7dd9ff)
 * 2014-12-04, SRS 2.0.51,  2.5k(2500) publishers, 91%CPU, 259MB. [commit](https://github.com/winlinvip/simple-rtmp-server/commit/f57801eb46c16755b173984b915a4166922df6a6)
+* 2014-12-04, SRS 2.0.52,  4.0k(4000) publishers, 80%CPU, 331MB. (mr_sleep=2000) [commit](https://github.com/winlinvip/simple-rtmp-server/commit/5589b13d2e216b91f97afb78ee0c011b2fccf7da)
 
 ## Architecture
 

@@ -165,17 +165,23 @@ int SrsFastBuffer::grow(ISrsBufferReader* reader, int required_size)
     }
 
     // when read payload and need to grow, reset buffer.
-    if (end - p < required_size && required_size > SRS_RTMP_MAX_MESSAGE_HEADER) {
+    // or there is no space to read.
+    int max_to_read = buffer + nb_buffer - end;
+    if (end - p < required_size 
+        && (required_size > SRS_RTMP_MAX_MESSAGE_HEADER || max_to_read < required_size)
+    ) {
         int nb_cap = end - p;
         srs_verbose("move fast buffer %d bytes", nb_cap);
-        buffer = (char*)memmove(buffer, p, nb_cap);
-        p = buffer;
-        end = p + nb_cap;
+        if (nb_cap < nb_buffer) {
+            buffer = (char*)memmove(buffer, p, nb_cap);
+            p = buffer;
+            end = p + nb_cap;
+        }
     }
 
     while (end - p < required_size) {
         // the max to read is the left bytes.
-        size_t max_to_read = buffer + nb_buffer - end;
+        max_to_read = buffer + nb_buffer - end;
         
         if (max_to_read <= 0) {
             ret = ERROR_RTMP_BUFFER_OVERFLOW;