Wednesday, April 18, 2018

Unetlab + CSR1000v performance

I've been using the following lab hardware for Unetlab for quite some time:

  • Intel(R) Xeon(R) CPU X5675 @ 3.07GHz (12 logical processors)

  • 96GB of ECC RAM

  • VMware ESXI 6.0.0 with vCenter 6.5 to manage things via the vSphere Web Client

  • Unetlab with images for IOL, CSR1000v, XRv, with all 12 cores dedicated to the VM, as well as 64GB of memory. Images used: csr1000v-universalk9.03.17.00.S.156-1.S and xrv-k9-6.0.1

I had been using 10 x IOL for most lab material, but decided to build-out a 10 x CSR1000v + 4 x XRv lab recently.

When utilizing this lab topology, carefully allowing everything to start and settle, I seem to get near 100% CPU utilization when doing just about anything on the CSR1000v's (memory never exceeds 60% utilization, and I have the page allocation for guests turned off as recommended):

  • Config replace can take 2 - 5 minutes per node. I get CPU watchdog log messages on each CSR1000v. The XRv's either don't complain or are fine.

  • Any kind of multi-console input to the CSR1000v's drives CPU to 100% (hitting enter on the console and having it send to all nodes for instance).

I've read quite a few entries from fellow labbers that they can run a 10 x CSR1000v + 4 x XRv lab without much trouble with far less of a hardware footprint (people using 2009 Nehalem architecture, as opposed to my beefier 2011 CPUs). The versions of code I'm running for the CSR1000v and XRv are in alignment with claims of a smooth experience.

Is there something I'm doing incorrectly, or is it really that the CSR1000v's can't run in tandem without blowing up the CPU? I don't want to upgrade Unetlab to EVE-NG just to find out that what I'm doing is folly to begin with.

As an addition, is there much value to running the CSR1000v's IOS-XE versus IOL from a critical feature standpoint?



No comments:

Post a Comment