NDA on AMD’s new FX CPU series “Bulldozer” was lifted yesterday and a lot of reviews are now available in the Internet. It looks like the new CPU architecture has its potential but it was released unfinished. Single threaded execution is lagging behind AMD’s old CPU architecture and the sheer size of the chip looks like a bad deal for AMD. Bigger chips are more expensive to manufacture and therefore AMD will have hard time to compete with Intel in pricing. Power consumption seems to be on a par with Intel i7 and i5 when in idle but jumps as high as 230W while Intel processors stay around 132W. This means more heat and more noise which is always a bad thing.
I was hoping fresh competition from AMD with their new architecture but it looks like they got a miss. Hopefully issues will be fixed and AMD’s new flagship shall shine. There is always hope..
Data scrubbing may render your RAID array unsable silently. To prevent that it is recommended that data concistency and bad block check is run regularly on RAID. This can be done by appending word check into /sys/block/mdX/md/sync_action, where X is your RAID device number.
echo check >> /sys/block/mdX/md/sync_action
I was pretty shocked when I noticed that the check was going to take 15000 minutes which equals to 250 hours. I planned to run this check bi-weekly and with these numbers my RAID would have been busy checking bad blocks 75% of time. It seemed like something was not right in the RAID. Luckily there is lot of information about Linux software RAID and their tweaking in Internet. Couple tiny changes into /etc/rc.local fixed the issue and now we are talking about magnitude of 360 minutes instead of 15000. Here are the fixes I made.
# This one alone helped the most
echo 4096 > /sys/block/md0/md/stripe_cache_size
# These two rows help with the read speed
blockdev –setra 4096 /dev/sd[a-d]
blockdev –setra 16384 /dev/md0