Notes on running a 3-node MapR cluster with Ubuntu 14.04 on Intel NUC6i7KYK (Skull Canyon).

Here is the hardware what I used to setup the NUCs:

• Intel NUC NUC6i7KYK Mini PC - 6th generation Intel® Core™ i7-6770HQ processor with Intel® Iris™ Pro graphics (2.6 to 3.5 GHz Turbo, Quad Core, 6 MB Cache, 45W TDP). ($570)
• Crucial MX300 525GB SSD - 525 GB SSD in Micron's MX-series product line. I/O speed advertised as 530 MB/s read and 510 MB/s write. ($150)
• Crucial 32GB (2x 16GB) SODIMM - DDR4 SODIMM laptop memory from Micron. DDR4 2400MHz PC4-19200 SODIMM 2Rx8 CL16 1.2v Notebook RAM. ($223)

• USB Keyboard
• 8-port ethernet switch
• Ethernet cables
• Stand-alone monitor with HDMI input
• Mini DisplayPort (Thunderbolt) to HDMI adapter
• HDMI display cable
• 2GB or larger USB stick

Download the Ubuntu 14.04 64-bit PC (AMD64) server install image from http://releases.ubuntu.com/trusty/, then create a bootable USB stick like this:

hdiutil convert -format UDRW -o ubuntu-14.04.5-server-amd64.img ubuntu-14.04.5-server-amd64.iso
diskutil list
# Then insert the USB stick
diskutil list
#Then identify the usb storage device, e.g. /dev/disk3
diskutil unmountDisk /dev/disk3
sudo dd if=ubuntu-14.04.5-server-amd64.img.dmg of=/dev/rdisk3 bs=1m
diskutil eject /dev/disk

Put the usb stick in the NUC and power it up. Go through the Ubuntu installation.

Create partitions for the Ubuntu operating system, swap space, and an unmounted partition for the MapR filesystem. Here's the partition table shown by running lsblk -o "NAME,SIZE,FSTYPE,MOUNTPOINT"

NAME     SIZE FSTYPE MOUNTPOINT
sda    489.1G
├─sda1 139.7G        /
├─sda2   9.3G        [SWAP]
└─sda3   340G

I assigned static IPs to each NUC with a config that looks like this:

I used hostnames, "nodea", "nodeb", and "nodec", and assigned them static IPs 192.168.2.4, 192.168.2.5, and 192.168.2.6.

auto em1
#iface em1 inet dhcp
iface em1 inet static
address 192.168.2.4
netmask 255.255.255.0
gateway 192.168.2.1

I'm primarily run these NUCs offline, but when I need them to reach the internet I connect them to my LAN and point them to a nameserver like this:

###/etc/resolv.conf

nameserver 192.168.0.1

Note, that will only work temporarily. The resolv.conf file will be overwritten when the NUC reboots.

Download and run the MapR installer like this on nodea:

wget http://package.mapr.com/releases/installer/mapr-setup.sh -P /tmp
sudo bash /tmp/mapr-setup.sh

Specify /dev/sda3 as the installation path when prompted.

On nodeb and nodec, create a mount point for MapR-FS with sudo mkdir /mapr and add the following to /etc/fstab:

nodea:/mapr   /mapr    nfs    auto    0   0

My applications ran a lot faster on these Intel NUCs than I expected. So I ran a few quick-and-dirty benchmarks to compare them with a MapR cluster I'm running in Azure and another I'm running on baremetal in a lab.

Running cat /proc/cpuinfo | grep bogomip:

• Skull Canyon: 41,472 BogoMIPS
• Azure DS14: 70,393 BogoMIPS
• HP Proliant DL380 G6: 89,600 BogoMIPS

Running hdparm -tT /dev/...:

That I/O result is surprising. Even though Azure provides SSD storage with the DS14, they only get spinning-disk level throughput. Combine that with the fact that virtual networks in Azure are generally limited to 1 gig ethernet speeds, makes me much more inclined to use the NUC for cluster computing.

Running mapr perfproducer -path /user/mapr/iantest -ntopics 1 -npart 3 -nmsgs 1000000 -msgsz 100:

Not too shaby! Given that I can buy an Intel i7 NUC for the same amount it costs me to run a DS14 VM in Azure for one month, I'd say these NUCs are pretty cost effective.