Hi All,
For any of you who are considering modifying starcluster to use the c5 (or
m5) instance types, I'll just highlight some caveats that are already out
there in the vastness of AWS's info:
There are no SSD drives available in these instance types, in case you have
been configuring them as extra storage (e.g., for localscratch or swap).
Also, block devices are now presented as NVMe devices. I happen to have
several EBS volumes in my /etc/fstab for my compute host AMIs, so cluster
launch was hanging because the compute hosts were being brought up in
emergency mode (which was viewable only after many minutes of waiting, thru
the "Get system log" option in the AWS console).
The device name mapping is described somewhat vaguely here
<
https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/nvme-ebs-volumes.html>.
The specifics fstab changes that worked for me: I updated /dev/sdb to
/dev/nvme1n1, and /dev/sdc to /dev/nvme2n1. (You can edit your fstab on a
c3 (or c4) imagehost per the above, and save the AMI.) Once you've modified
starcluster static.py to include the new instance types and reinstalled,
your launch should succeed.
(Note also that your image must also be configured to support ENA, and your
OS must also have ENA support enable, as described here
<
https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/enhanced-networking-ena.html>
.)
Fyi,
Lyn
Received on Mon Apr 02 2018 - 14:16:23 EDT