Virtualization has changed the face and footprint of IT, but one fact remains – we still demand optimum performance and value from our IT investments. This article takes a closer look at one hypervisor option – Microsoft’s Hyper-V – and what generation of the virtual server is best to maximize IBM Domino server’s performance.

For my investigation, I utilized the following setup:

Microsoft Server 2012 R2 with Hyper-V Role Installed

4 Windows Virtual Machines running IBM Domino 9.0.1 FP5

     Generation 1 (4GB RAM / 2 Virtual CPU Cores / 60GB HDD)

  • Windows Server 2012 R2
  • Windows Server 2016 TP5

     Generation 2 (4GB RAM / 2 Virtual CPU Cores / 60GB HDD)

  • Windows Server 2012 R2
  • Windows Server 2016 TP5

1 Windows 10 Virtual Machine with Lotus Notes 8.5.3 Installed (4GB RAM / 2 Virtual CPU Cores)

Depending on your familiarity with Hyper-V, the term ‘Generation’ may or may not ring any bells. For those who haven’t had the chance to get their hands on this hypervisor, let me dive a bit deeper.

As Hyper-V has evolved, Microsoft has continued to tweak their virtual machine model. The older model, referred to as Generation 1, included emulated legacy drivers such as floppy drive controllers and has backward compatibility to Hyper-V’s beginnings on Windows Server 2008. Generation 2 virtual machines are much sleeker, having eliminated the majority of the emulated controllers and hardware devices. The prevailing thought is that this, along with numerous other changes, translates to better overall VM performance. My goal here is to gauge whether or not this better performance shines through on VMs running IBM Domino server workloads.

For those who’d like to read Microsoft’s full TechNet article on Gen 1 vs Gen 2 virtual machines, I invite you to head on over to https://technet.microsoft.com/library/dn770158(ws.11).aspx .

Once my Domino servers were up and running, I had to track down a good tool to simulate load. To meet this need, I dug up IBM’s Server.load tool which is designed exactly for this type of testing and performance tweaking.

White Paper- http://public.dhe.ibm.com/software/dw/lotus/DominoServerLoadFinal.pdf

Once the necessary agents were deployed within the Domino environment, I was having a hard time getting the tool (which is actually installed by default along with Notes Administrator) to run alongside my local Notes 9.0.1 client. After trying it on various Windows 7, 8, and 10 workstations, I finally threw in the towel and installed the older Lotus Notes 8.5.3 client — It had been a few years! Thankfully, the tool’s executable — sload.exe — launched without issue with this version of Notes and I was free to test using any of the built-in testing scrips:

Note: In an effort to get the most accurate results, I tested the servers one by one (with the others in a shutdown state) so as to ensure resource contention did not skew my findings. Additionally, I terminated any Domino services not required by the Server.Load scripts (replicator, indexer, etc.)

After following IBM’s deployment checklist, I decided a good first test was gauging individual server response times during the creation of standard mail databases, 10 of them per server specifically.

The N85Mail8 Initialization script, built into the Server.Load tool, completed on all 4 servers in very similar timeframes. However, I was surprised to note that average server responsiveness (in milliseconds) was faster on the Generation 1 VMs!

With valid mail databases in place, I put the servers through 5 iterations of the mail routing event script (R6 Mail Routing) which simulates the simultaneous use of each mail database. This included view refreshes, type ahead lookups, and more:

As with the mail initialization script, the 2012 R2 Generation 1 virtual machines once again took the top spot in fast response time. Also notable is that the Windows Server 2012 servers were more responsive compared to their Server 2016 counterparts for this test.

Last but not least, I wanted to put some decent HTTP load on each of the servers through utilization of the DWA (iNotes) workload script. Using this script, I simulated 20 iNotes users which generated much higher response times in general compared to the other scripts:

As visible in my results above, the fastest server ended up being the 2012 R2 Generation 1 virtual machine once again, though the Generation 2 Server-Win4 was not far behind.

In summation, the data collected would suggest that virtual machine generation, at least on the Hyper-V platform, does not have a large impact on overall IBM Domino server performance and response times, though specifically in smaller to medium size environments. The performance gains very well may shine through on large-scale deployments with hundreds of users and mail files or on Linux operating systems such as RHEL or Ubuntu. Only testing and data will show for sure – so stay tuned!