On Friday 19 June 2009 16:40:43 Cathey, Jim wrote: > "cat $FILELIST > /dev/null &" will prepopulate your cache for you, and > should > do it on a separate processor from the md5sum process. (I take it this > is > jffs2 log reassembly and decompression?) > > Yes. It took 11.5 seconds to cat every file in the list to /dev/null, > so in fact it's a net loss over the -j2 option to md5sum!
Even backgrounded? > The second > catting took 0.2 seconds, obviously the big loss is in jffs2. The point is to keep one processor busy doing jffs2 decompression work while the other processor is doing md5 stuff. (You mentioned it was an SMP system.) If you serialize the two of them, yeah it's likely to be a net loss. The cat by itself should be trivial (as you noticed, 0.2 seconds out of cache), it's just there to let jffs2 work happen in parallel with md5 work. But yeah, sounds like the dominant factor here is cache population in the kernel, and most of the CPU is consumed by the jffs2 driver. It would make just as much sense to add a -j option to cat as it would to add it to md5sum here. > -- Jim Rob -- Latency is more important than throughput. It's that simple. - Linus Torvalds _______________________________________________ busybox mailing list [email protected] http://lists.busybox.net/mailman/listinfo/busybox
