Use the VMXNET3 drivers for your virtual interfaces, otherwise everything is pretty straight forward. I would assign 2 cores to the machine and 4-8gb of memory, maybe more if you have a lot of users (probably not for home). I use 2 cores/4gb of memory of mine with most features except for email enabled with a 150/20 internet connection.
Hi Jim,
There aren't any specific guides to how to setup a UTM on ESXi but I can give you a few pointers. What hardware do you have for the host machine by chance?
Firstly, don't use the prebuilt vmware vm's. They are built to be as compatible with ESX as possible so use extremely conservative defaults. Best is to create your own VM and install from the software ISO (not the appliance/SSI one).
Guest wise, use either the Generic Linux 64 bit guest or the SUSE Enterprise 10 (SLES) profile. Either one works. For hardware I always use the SAS option for the vdisk typically 40 GB thick provisioned, and VMXNET3 hardware for the vNIC's. For home use I normally suggest 2 vCPU w/ 6GB RAM.
That config on my Intel i7-9x0 host machine handles two people (one heavy, one light user) on a 100Mbps connection no problem. Have IPS, Web Filtering, Single Scan A/V, etc and it runs fine.
That's some serious CPU allocation man. [:)] I'm assuming that guest does transcoding?
One thing to watch out for when using multi vCPU configs in ESXi, scheduling contention can become a performance issue. Back in the 4.x days VMware did some studies which found on hosts running mixed single vCPU, dual vCPU & quad vCPU guests, the dual vCPU guests often outperformed the quad guests in a lot of benchmarks. Reason was ESXi schedules CPU time slices for all vCPU cores simultaneously which means it has to have four physical cores free to schedule a 4 vCPU guest whereas a 1 or 2 vCPU guest only need 1 or 2 physical cores. Long and short was that single & dual core guests had less contention for CPU time meaning they got a better slice of the pie, resulting in better performance then their quad brethern.
And given your allocating 6 vCPU into a 4 core Xeon (8 w/ HT but not true cores), you'll want to look at CPU ready values. This site has a good way to analyze the data from the CPU graphs: CPU Ready Revisted - Quick Reference Charts - VMtoday
Interesting, that's not something I was aware of. Yes my server does a lot of transcoding, sometimes 3-4 files simultaneously. I wonder what would be the best way for me to test the performance of allocating 2 vs 4 vs 6 vCPU.
