You Did WHAT??? - Part 1

Or, Once Upon a Time in Cobbler

Infinite, Cosmic Power!

When I began designing the SLES for SAP v15 environment for HANA 2.0, my initial brief was for it to be hosted almost exclusively on a brand-new IBM Power9 infrastructure. This green field was based on E980s, 80 cores and 16TB of RAM each, connected to brand-new PowerMAX frames with nearly 1 Petabyte of storage. Did I mention the 10Gb network infrastructure. The 6 Broadcom NICs per E980? SMT-8 and top-of-the-line RAM? Multiple brand-new HMCs?

If the infrastructure was any faster, it would have created a causal loop and we would have been writing data before we actually had it.

OK, I’m joking. A little.

What Fine Print?

As we got into planning in earnest, we found out that our specific version of NetWeaver had issues with PowerPC (I understand this has since been solved). The organizational pivot, late in the design stage, meant that the requirements grew to include a substantial population of VMware-based VMs.

I determined the organization needed a unified system build infrastructure, one that supported deployment of both x86_64 VMs and ppc64le LPARs. Obviously, SuSE Manager would be central to this design - but it would have to dynamically handle builds of both architectures; early plans called for over a hundred HANA LPARs alone, and I didn't want to hand-craft an AutoYaST for each one individually.

Surely I couldn't be the first to do this, could I? After some looking around the web for example implementations, I realized there was a lack of any relevant online discussions. I decided to let our SAP Architect know my plans, and he promised he'd ask around. After all, he'd been an SAP employee and knew people who were already implementing HANA.

A few days later, he said he'd made some calls and talked to some friends in Germany, people who knew SuSE. Their reply to my plan? No, they didn't know anyone who'd created a unified build infrastructure that simultaneously supported x86_64 and ppc64le deployments. Not only that, they doubted it could be done.

Failure Is Not… Well, You Know the Rest

Personally, I thought it was quite doable. With not much more than 8 months until Go Live, faced with dozens of LPARs and VMs to build, I decided to see who was right (Spoiler: It was me)

Starting with this entry, and continuing over the coming weeks, I'll lay out my implementation of a SuSE Manager environment designed to deploy both x86_64 VMs and ppc64le LPARs; and, how I engineered an OS build process that is as close to identical as possible for each (and why I found that helpful).

And So It Begins...

The first thing to do is install SuSE Manager (SUMA). I installed v3.2 (which was the latest at the time), but this should be applicable to v4.x. SuSE has published a guide focussed on SuSE Manager and PowerVM. While dated, the documentation is a reasonable step-by-step to get you started.

It is important to understand that a SuSE Manager running on x86_64 is perfectly
capable of deploying to PowerPC LPARs (and vice-versa). The platform of the SuSE Manager is not important in that regard (indeed, my experience is that while SuSE will sell platform-specific SUMA entitlements, SCC doesn't actually care what chipset is used when the host is registered).

Except for using an x86_64 VM, I followed that guide, as appropriate SUMA v3.2, through Page 35/Figure 49. My design started to diverge at that point. In future entries, I will explain, in full, how I delivered the final design, including technical details and commented sample configuration files. So far, I've planned for this series of entries to include (titles and ordering subject to change):

  • Images vs. Configurations (which may be of particular interest for organizations not yet ready to embrace SALT and Docker)

  • No, you really don’t want to use SLES v12

  • Designing a Build network

  • Studies in SR-IOV: Is minimizing VIO worth the LPM impact?

  • DHCP server configuration to support VMware vs PowerVM

  • An analysis of PowerVM, its use of BOOTP, and how that impacts building SLES for SAP

  • How to implement a DNS server to ensure smooth registration of hosts

  • What you really need from the bootstrap script (and how it actually works)

  • A workable SLES for SAP AutoYaST configuration (and how to modify it)

  • XFS Uber Alles - Avoiding Multiple Filesystem Types

  • The Tao of udev on PowerPC

  • Storage Management for HANA

  • Rescue Mode for PowerPC (Avoid the VIO-d)

And probably quite a few more.

So please, subscribe

and stay tuned…