Support and News Forum

hi

xrv9k node is very heavy...
so,when some nodes works, a node is downed.
are there any measures?

*xr9k node seems to be downed when the cpu utilization of the hypervisor becomes 100%.

thanks

What are the spec of your eve server ? CPU assigned, type of cpu, memory ???
E.

hi

the spec of eve server are as follows.

Xeon 24 CPUs
memory 64 GB

Which Xeon ?

Old xeons are weak and a simple i7 can achieve a better job...
XRv9K use 4 vcpu and 16G meaning you could only start 4 XRv9K without uksm.
I You use uksm, you should not start all of them at once.... but one by one...

E.

Yes, they are. with a few thousand routes usage hovers around 12G.
You will need to decide which features you need. If you can get buy with running xrv <= 6.2.3, the load will be less.
If you however need the features of say 6.4.1 (l2vpn signaling and payload for example) and run a multiple node lab of decent size then you need to decide how much server you can devote to your lab.
If you happen to be simulating a carrier environment then I have had adequate performance running bare metal. You can make the server multiple boot so not to lose your esxi environment for other projects.


what does lscpu show ?
Prevent swap with added ram ?
Are you on SSD?
If not do not pass go. Proceed to microcenter and buy and sata SSD or NVME. ( you may not be able boot from NVME but you can run /opt from the NVME)
Insure your running the 19 November release with CPU settings improvement for qemu nodes ; however you can set this with sysctl for any of the versions
Consider running bare metal if you have a long term need for a lab such and pre-deployment checking and you budget allows for it.
Place it in a friends data center with an iLOM or remote outlet management to save power.
Lease VPUs from the cloud.

I can run a max of about 30 xrv9ks qemu VM ( 6.5.3) and maintain decent response times on a Dell R810 with 4 x E7-4870s + 128G of ram and 256G of ram ( running bare metal ) should you want to disable uksm in order to keep the memory access local to a processor.
If you have more VMs than this you can split them between multiple servers and connect them with a pnet# cloud ( broadcast domain ) using sub-ints with unique vlan numbers. be sure not to duplicate mac addrs if using this method . If you live in a cool climate and do not have access to a date center you can heat your apartment with this setup.

hi

thanks for the detailed explanation !

i'll consider in various ways.