By DJ Spry, Apstra
It is a radical decision, and it might seem like a step into the unknown – were it not for the online interactive Network Automation OpEx Calculator. You enter your own estimates for each step in designing, building and operating a leaf-spine data center architecture, and receive a detailed analysis of your likely savings.
The results are impressive. Typical users would average around 83% greater efficiency:
- Saving 23.94% for Scoping Requirements
- 72.99% at the Design Stage
- 82.26% during Implementation and Test
- 75.00% during On-Going Operations.
These figures are impressive, and would justify automation for any new data centre network – but what about legacy networks? Can an existing network be upgraded for greater automation and achieve that 75% OpEx saving?
The legacy headache of NOS upgrades
Upgrading network operating systems (NOS) once meant replacing loads of PROM chips. It was excruciating: but at least NOS software was simpler in those days, and these upgrades were seldom required.
Network devices now have flash memory to hold several NOS images, making it much easier to “back out” a NOS upgrade if the new version does not work as expected. But the process still involves fragile manual processes plus tedious and redundant testing:
- Secure shell to device.
- Secure copy NOS image to device.
- Change and save boot configuration.
- Gracefully drain traffic.
- Reload.
- Watch console and cross fingers.
- Validate what was broken – and service is restored.
Although spared the hardware, the increased complexity of NOS software means that upgrades are still a headache – for example when loading an upgrade to fix bugs, security vulnerabilities or other issues.
A final straw can be the finger pointing and blame games whenever “issues” arise. Question the software, and you will be told it is a hardware issue – and vice versa. Finally, the finger points back to the operator, who surely must have misread the instructions, or has simply done something wrong. Operating costs soar while egos are defended.
Network operators are crying out for a simple process guaranteed to manage simultaneous upgrades as well as upgrades across multiple vendor types – bearing in mind that vendors apply a range of procedures (POAP, ZTP, ONIE) for the upgrade/downgrade process.
Proper automation manages such diversity
The aim of total management software should be a single platform to automate the full network lifecycle, so that any network will be able to configure, repair and validate itself automatically – or with minimal human intervention.
With each new version, Engineering Testing and Global Support Teams should select up to three recommended versions of each NOS and rigorously test that each works consistently against a Reference Design for data center, spine-leaf networks. Keeping fully up-to-date could entail fifteen million tests a day.
Once the NOS has passed these tests, you might provide “Certification” for that specific combination. It means customers can be confident that the software will automatically align any of the certified operating systems with all the vendors’ equipment without any need for manual intervention.
Should the tests reveal any new failures or regressions in a NOS, the vendor would be contacted, to ensure the problem will be fixed. More sophisticated management software should also detect potential issues affecting possible deployment combinations and alert or, ideally, proactively correct incompatibility problems.
For example: an issue exists with Cumulus Linux, where on certain network hardware ASIC models (e.g. Broadcom Trident II+), in certain network architectures using VXLAN encapsulation, the network switch does not properly rewrite network addresses and their time-to-live (TTL) values. For any customer with Cumulus Linux devices installed, the best management software can compare the Cumulus Linux version against the hardware, determine if the workaround is needed, and then automatically implement the adjusted configuration without any effort by the user.
If a network vendor recommends that the customer uses a different NOS version, the customer can contact Global Support to ask for testing and support for the new version. After a successful upgrade, Intent-Based Analytics (IBA) informs the user when the device is online and ready for operation.
Less hassle. Less OpEx.
A system that automatically aligns best-practice reference design with recommended vendor NOS versions across a multi-vendor network, bypasses traditional Network Operators’ headaches. Their work becomes less frustrating. They have more freedom to think and plan. The business runs more efficiently, with fewer complaints and a dramatic saving in operating costs.
Reason enough to explain why so many Fortune 500 companies are already deploying this automation solution.
