[Nerdsniped] use significant parallelism for warmup IO

pde · pde · commit 34ecd5d84eb4 · 2015-04-23T15:05:36.000-07:00
- This will let the OS disk scheduler seek efficiently
diff --git a/makexpi.sh b/makexpi.sh
@@ -1,4 +1,4 @@
-#!/bin/sh
+#!/bin/bash
 set -o errexit
 APP_NAME=https-everywhere
 
@@ -52,11 +52,15 @@ if [ -n "$1" ] && [ "$2" != "--no-recurse" ] && [ "$1" != "--fast" ] ; then
 fi
 
 if [ "$1" != "--fast" -o ! -f "$RULESETS_SQLITE" ] ; then
+
   # This is an optimization to get the OS reading the rulesets into RAM ASAP;
-  # it's useful on machines with slow disk seek times; there might be something
-  # better (vmtouch? readahead?) that tells the IO subsystem to read the files
-  # in whatever order it wants...
-  nohup cat src/chrome/content/rules/*.xml >/dev/null 2>/dev/null &
+  # it's useful on machines with slow disk seek times; doing several of these
+  # at once allows the IO subsystem to seek more efficiently.  Note that
+  # bash uses the completely insane glob ordering aAbBcC...
+  for firstchar in `echo {a..z} {A..Z} {0..9}` ; do
+    # Those cover everything but it wouldn't matter if they didn't
+    nohup cat src/chrome/content/rules/"$firstchar"*.xml >/dev/null 2>/dev/null &
+  done
 
   echo "Generating sqlite DB"
   python2.7 ./utils/make-sqlite.py
