One of the major tech books to hit shelves this year is the VMware vSphere 6.5 Host Resources Deep Dive. The title is quite a mouthful! I had the pleasure of chatting with the co-authors of this book, Frank Denneman and Niels Hagoort, during a nerdy session of the Datanauts Podcast entitled Diving Deep Into vSphere Host Resources. Since then, I’ve had many more opportunities to return to the book as a go-to reference guide for compute, memory, storage, and network information. In this post, I’ll highlight some of my favorite bits and bytes within the book in hopes of expressing why I like it so much.
Update! If you want to download a FREE copy of the book, Rubrik is giving them away here!
Full Disclaimer: I snagged a signed copy of the book while at VMworld Europe as part of Rubrik’s campaign to freely distribute over 2,000 copies to attendees.
Highlights from the Book
The book is sliced up into four major parts – compute, memory, storage, and network. Each part has an interesting, if not anecdotal, prologue produced by someone dazzling in the industry, such as Pat Gelsinger on compute and Carl Waldspurger on memory. I found this layout to be great for using the book as a reference guide and have planted little colored flags into each section to remind me where the juicy bits lay.
The compute part that is focused on NUMA (Non-Uniform Memory Access) is particularly dear to my heart. During my consulting days, I spent time at technical workshops deep diving into how the architecture of CPUs and memory banks influenced NUMA, vNUMA, and performance of workloads. This is particularly important for VBCAs (Virtual Business Critical Applications) riding on so called “Monster” VMs. The Host Resources Deep Dive book spends a delightfully long time going into the interaction between vmkernel, board, virtual machine, and other aspects of NUMA, in addition to scheduling considerations. In itself, this part of the book alone is worth the price of admission.
If you’re curious on vmkernel memory management, chapter 11 in the memory part is right up your alley. I’ve nearly emptied an entire highlighter on page table walks, MMU, SLAT, Translation Lookaside Buffer (TLB), and other fun, geeky topics. The peppering of “worse case scenarios” is pretty groovy and highlights the impact of stressed architectures. I will also give an honorable mention to the discussion on disabling large pages since that is commonly a point of contention for server and virtualization administrators looking to squeeze every drop of memory out of a hypervisor host.
From a storage part perspective, my favorite reading was found in chapter 14 on non-volatile memory architecture. I recently recorded a Datanauts episode with Dr. J Metz entitled “NVMe’s Impact To The Network” (publishes later this month) that went super deep into the world of NVMe and NVMe over Fabrics. Thus, having this book as a primer to the world of insane throughput and IOPS was a crux to my discussion. Additionally, the segments describing reads, writes, and read-modify-writes to flash are critical for those looking to up-level their knowledge from spinning disk up to cell based architectures – because it’s completely different! The hosts do a wonderful job explaining wear leveling, write amplification, bad block management, and TRIM.
The end of the book contains the network part of the deep dive. As co-author of the Networking for VMware Administrators book, I was keen to see what content made the cut. While the hardware discussion is a solid primer for those looking to understand the pluses and minuses presented with PCIe, CNA, and other network adapters, my nerd propeller turned at the vSwitch Uplinks details found in chapter 21. This is often the area most misunderstood by virtualization engineers, especially with LAG (Link Aggregation) technologies are brought to bare. Make sure to grok all of the LACP, LBT (Load Based Teaming), and hashing techniques found in the book.
This book is not light reading. You won’t just pour over it in a few nights and walk away saying “OK, great, I now know everything about host compute, memory, storage, and networks.” Expect to revisit it time and time again as a piece of evergreen content. And that’s great! Frank and Niels have taken a large quantity of fragmented tribal knowledge and combined it into one jam-packed book for the masses.