Terrible timekeeping with trusty on EC2

[Ross Vandegrift]

The official trusty EC2 AMIs exhibit terrible clock jumps. I've included some syslog timestamps below that show the problem.

I believe it's a kernel issue - switching to the kernel linux-virtual-lts-xenial fixes it.

A few things I've tried that don't help:
- running ntpd with low-stratum pool servers
- disabling trusty's dhcp-triggered ntp stop (see https://bugs.launchpad.net/ubuntu/+source/ntp/+bug/1278359/comments/7)
- switching the system clocksource to tsc from xen pv

This is ami-8cda19e1 in us-east-1. The problem has affected many builds though. I don't think it's related to instance type (I think I've seen it on t2, c3, c4, and m4).

$ zcat /var/log/syslog.3.gz | awk '{ print $1" "$2" "$3 }'
...
Sep 16 07:55:29
Sep 16 07:55:29
Sep 16 07:55:29
Sep 15 18:34:01
Sep 16 07:56:01
Sep 15 18:35:01
Sep 16 07:57:01
Sep 15 18:36:01
Sep 16 07:58:01
Sep 15 18:37:01
Sep 16 07:59:01
Sep 15 18:38:01
Sep 16 08:00:01
Sep 15 18:39:01
Sep 16 08:01:01
Sep 15 18:40:01
Sep 16 08:02:01
Sep 16 08:15:01
Sep 16 08:17:01
Sep 16 08:17:01
Sep 12 03:51:01
Sep 16 08:18:01
Sep 16 08:19:01
Sep 16 08:20:01
Sep 16 08:21:01
Sep 16 08:24:00
Sep 16 08:24:00
Sep 16 08:24:00
Sep 16 08:24:01
Sep 16 08:25:01
Sep 16 08:26:01
Sep 12 04:05:01
Sep 12 04:06:01
Sep 12 04:07:01
Sep 12 04:08:01
Sep 12 04:09:01
Sep 12 04:10:01
Sep 12 04:11:01
Sep 16 08:38:01
Sep 12 04:13:01
Sep 16 08:40:01
Sep 16 08:41:01
...