Some thoughts on speed up firefox start time

This is common issue, it's getting critical when the kernel takes 10 seconds to mount root, but firefox spends around 40 seconds to make a first start.

If we use a simpler XUL interface and xulrunner instead of original firefox, it can save about 9 seconds. (Maybe that's why the Prism project exists :-)

After profiling we can find out the problem is caused by a huge amount of I/O, the first start of xulrunner will read nearly 75MB files including .so, .xpt and fonts. (See the whole files list: readahead.txt )

Survey

Here's some common approach:

• UPX - binary compression
  
• prelink - static library
  
• readahead - prefetching files
  
• swiftweasel - CPU optimization

So I decide to try on prefetching, by using simple command: cat `cat readahead.txt` >> /dev/null .

Result

Sorted by total time:

	mount	prefetch	xulrunner	total time
no-prefetch	10s	N/A	28s	40s
parallel-prefetch	10s	23s		44s
prefetch	10s	16s	15s	46s
toram-prefetch	20s	14s	15s	51s

You may notice there's no mechanism to ensure reusing of cached file, so the prefetching total times is longer, although the start time is reduced.

Conclusion

1. file priority or size should be taken into account when prefetching
   
2. the next bottleneck target is the start script (xulrunner-1.9 or run-mozilla.sh), it takes 7 seconds
   
3. standalone firefox?