Changeset 54ef438

Timestamp:

12/10/18 11:14:57 (2 years ago)

Author:

Shane Alcock <salcock@…>

Branches:

develop

Children:

24fe0f5, 509ee47, c2fe7b6

Parents:

c337c9f9

Message:

ndag: just read one "batch" of packets per pread call

Previously, we would stay in pread until we got close to the
full number of packets requested or a non-blocking read gave us
no new packets.

This meant that we could in theory waste time doing an unnecessary
non-blocking read that was going to fail, when instead we could
just assume that we were going to be called again shortly anyway
if the user wanted more packets.

It's a bit tenuous when I try to explain it, but it seems to save
us a bit of CPU time in practice...

File:

• lib/format_ndag.c (modified) (4 diffs)

lib/format_ndag.c

@@ -1054 +1054 @@
         }
 
-        zerotv.tv_sec = 0;
-        zerotv.tv_usec = 0;
-
         for (i = 0; i < rt->sourcecount; i++) {
                 if (rt->sources[i].sock == -1) {
@@ -1088 +1085 @@
         }
 
+        zerotv.tv_sec = 0;
+        zerotv.tv_usec = 0;
         if (select(maxfd + 1, &fds, NULL, NULL, &zerotv) == -1) {
                 /* log the error? XXX */
@@ -1259 +1258 @@
         rt = (recvstream_t *)t->format_data;
 
-
         do {
                 /* Only check for messages once per batch */
@@ -1265 +1263 @@
                         rem = receive_encap_records_block(libtrace, rt,
                                 packets[read_packets]);
-                } else {
-                        rem = receive_encap_records_nonblock(libtrace, rt,
-                                packets[read_packets]);
-                }
-
-                if (rem < 0) {
-                        return rem;
-                }
-
-                if (rem == 0) {
-                        break;
+                        if (rem < 0) {
+                                return rem;
+                        }
+
+                        if (rem == 0) {
+                                break;
+                        }
                 }