[clug] Which to go for?

Andrew Janke a.janke at gmail.com
Sat Jan 16 03:20:54 MST 2010


> Oh I know the SSD has much better bandwidth, I'm just wondering what real-world situations can stress this if pulling ~300MB of source files off disk in under a minute doesn't do it.  So yeah, if getting another few GB of RAM can make a 1TB spinning disk perform within 10% of a 30GB SSD in this workload, I'm stoked :-)

<tick!>

>> I do! :)  I like.  I use them for system disks on Cluster machines (a
>> 20GB /tmp predominately for temporary files).
>
> Hehe, I don't know I'd class Cluster machines as home/office/workstation but I see your point!

Well for me a "cluster" is little more than a bunch of desktop
machines in a room or two linked with a gig switch. You then stuff a
video card in a few of them and use them as workstations with "stuff"
happening in the background. So in a way they are "desktop machines".
;)  I see zero point in buying a new cluster and having a clunker on
your desk, the older machines then have their video cards yanked and
are retired to the corner to munch away at data for me.

On the kernel note, I just got bored and did this (on an i7 920, 12GB
RAM, 30GB OCZ vertex system "disk" machine running karmic koala
2.6.31-17-generic).

  # cd /tmp

  # wget http://www.kernel.org/pub/linux/kernel/v2.6/linux-2.6.32.3.tar.gz

  # tar zxvf linux-2.6.32.3.tar.gz

  # cd linux-2.6.32.3/

  # cp /boot/config-2.6.31-17-generic .config

  # time make -j6
   <...>
  IHEX    firmware/yam/9600.bin

   real	11m15.044s
   user	55m46.180s
   sys	5m47.400s

Note the 'make -j6', an i7 with HT isn't really 8 cores as we all know
and I find 6 concurrent jobs to be about the sweet spot. 7 is
sometimes a frazzle faster but 8 is nearly always slower if all
"cores" are running at 100%.

--
Andrew Janke
(a.janke at gmail.com || http://a.janke.googlepages.com/)
Canberra->Australia    +61 (402) 700 883


More information about the linux mailing list