NTP Skew or Smear Sources

Mixing algorithms is tricky

Andrew Latham

I promote redundancy at every chance. Having extra DNS servers or extra DHCP servers is always grand. One issue that I noticed recently was NTP Skew or Smear. This is a method of slowly adjusting the clocks on systems to compensate for leap seconds which happen from time to time. The process is well thought out and well documented however the implementation is not perfect. While discussing redundancy a network engineer was explaining the current solution and future plans. I was asked about the NTP setup and advised them to diversify vendors of hardware solutions. After a deeper dive I found they were using cloud provider vendor NTP services from a directive up the command chain. The issue came when I saw they were configuring multiple providers NTP servers as sources. The issue here is that NTP sources might use a different smear method and the inclusion of non-smeared sources would mean the NTP calculation would favor the lowest latency source as the various vendors smear drifted apart from atomic time.

Mixing algorithms is tricky

Andrew Latham

A recurring issue in most computer system infrastructure is the System of Record (SOR)[1] which is a source of truth about the desired state or the current state of systems and networks. Many outsiders or senior management are left to believe that the planned state and the current state are the same but that rarely happens. I have written a few SORs and enabled others to be more accurate. Much of the work on modern container image manifests [2] has made great advancements to a more complete language to communicate the details of an endpoint. An issue I would love to resolve is the assumption that a single application is the SOR. The reality is that a data service is the ultimately the SOR and many things might read from or write to it. Today if I was asked to write another SOR I would setup RethinkDB [3] and establish a table named 'help' that pointed to a table named 'standards' and thus became self documenting system.

SORs should also be free to access offline in emergencies and distributed systems like RethinkDB and even Git are perfect for this type of setup. A support person or team can maintain a local copy of the SOR with little to no overhead. Even tools like MediaWiki[4] are great as they can be duplicated or distributed as a side effect of their design. For discovery there are established tools like DNS SRV [5] that are often overlooked. A popular solution for datacentres is NetBox [6] which is a fine step forward from RackTables [7] which served many for years and is still not a horrible solution. In my spare time I hope to integrate a DCIM [8] solution into Odoo [9] to connect the various organizational groups together.

  1. https://en.wikipedia.org/wiki/System_of_record
  2. https://github.com/opencontainers/image-spec/blob/master/manifest.md
  3. https://www.rethinkdb.com/
  4. https://www.mediawiki.org/wiki/MediaWiki
  5. https://en.wikipedia.org/wiki/SRV_record
  6. https://netbox.readthedocs.io/en/latest/
  7. https://www.racktables.org/
  8. https://en.wikipedia.org/wiki/Data_center_infrastructure_management
  9. https://www.odoo.com/

Internet Authorship - Content Size and Frequency

Internet Authorship - Content Size and Frequency

Andrew Latham

While watching Rand talk about this I was looking for the option to instantly buy him a beverage of his choice. To often I see these reports on best practice and averages that aspire to address every type of content in every industry. Not all audiences are the same for all industries. I was thinking about this and a great comparison could be measurement conversion in cooking where a short concise answer that is less than 10 words is more useful for the audience versus content describing the history of cooking flour where we would expect longer content and even citations to more content both online and off.

Write a lot of content to become a better content creator. Publish content for the domain or site that your audience would find useful. Review your metrics always.