[c-nsp] CSR1000V and CPU usage

[Pshem Kowalczyk]

For completness, this is what we use to get higher performance out of
CSR1000v on ESX:

1. Set Latency Sensitve to High  (
http://www.vmware.com/files/pdf/techpaper/latency-sensitive-perf-vsphere55.pdf
)
2. ethernetX.coalescingScheme = "disable" (for all interfaces on the VM)
3. Pin down memory and CPU (we run a 4 vCPU setup with 4G of RAM)

This are the essentials from the config:
config.version = "8"
virtualHW.version = "10”
numvcpus = "4"
memSize = "4096“
ethernet0.virtualDev = "vmxnet3"
ethernet0.networkName = "VM Network"
ethernet0.addressType = "vpx”
ethernet0.present = "TRUE"
ethernet1.virtualDev = "vmxnet3"
ethernet1.networkName = "TG1"
ethernet1.addressType = "vpx”
ethernet1.present = "TRUE"
ethernet2.virtualDev = "vmxnet3"
ethernet2.networkName = "TG2"
ethernet2.addressType = "vpx”
ethernet2.present = "TRUE"
guestOS = "other26xlinux-64”
sched.cpu.htsharing = "internal"
sched.cpu.min = "13996"
sched.cpu.units = "mhz"
sched.cpu.shares = "high"
sched.mem.min = "4096"
sched.mem.shares = "high”
ched.cpu.latencySensitivity = "high"
sched.mem.pin = "TRUE"
ethernet1.coalescingScheme = "disabled"
ethernet2.coalescingScheme = "disabled”

(we use Gi1 for management (ethernet0 in ESX).

kind regards
Pshem


On Fri, 14 Aug 2015 at 01:06 Mark Tinka <mark.tinka at seacom.mu> wrote:

>
>
> On 13/Aug/15 14:07, Nick Cutting wrote:
>
> > There were some discussions on here a few weeks/months back about some
> best practices for running these devices.
>
> I think it was Pshem who shared some tweaks to ESXi to improve performance.
>
> >
> > Mostly folks were using these for Route reflectors I think.
>
> That's what we do. Works like a charm, over 12x months now.
>
> Mark.
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>