Clock Skew Exists

Re:Clock Skew Exists by packetized on Wednesday February 10, 2016 @09:28PM (#11077859)

Clock skew is a bad enough thing that it's accepted wisdom to not run timing-sensitive authentication authorities (read: Kerberos, mostly) in a hypervisor or VM container.

[ Parent | on HN ]

Re:Clock Skew Exists by josh2600 on Wednesday February 10, 2016 @10:12PM (#11078007)

This has a few of reasons:

1) all abstraction necessarily adds a performance penalty.

2) Hypervisors or VMs have core affinity, but, often, multiple VMs and Hypervisors are pinned to the same core. Since one cpu cannot expose time to two different vm requests at precisely the same time, there is always some variance.

3) Process isolation is pretty good, but total performance isolation is not quite the same.

There are a TON of papers written about VM and hypervisor time management. It's all fascinating and hard.

[ Parent | on HN ]

Re:Clock Skew Exists by packetized on Wednesday February 10, 2016 @10:43PM (#11078102)

Absolutely. I'm just surprised that this many people seem to be in the dark about clock synchronicity (or lack thereof) effects on distributed systems.

[ Parent | on HN ]