Creating a dynamic lab environment with vEOS and GNS3

[NetworkGroover]

Hey guys - I'm writing a guide on using Arista vEOS with GNS3 if you're interested.  I've got another whole section I'm going to add to it regarding creating dynamic environments (Run a single task to re-provision the entire network from L3 ECMP to L2 MLAG and back) with Ansible, but the two parts I have in it now should have some handy info if you're playing with this now or are considering it.

The content is in document format, but I've went ahead and slapped it into my blog in two different posts here:
http://aspiringnetworker.blogspot.com/2016/03/creating-dynamic-lab-environment-with.html

[routerdork]

Cool

[wintermute000]

anything particular re: GNS3 vs UNL? in Unetlab, I had a VSRX pinging a IOSv through a VEOS switch in around 15 minutes flat
Colleague of mine has an Arista deployment project (lucky bastards), he's labbing VEOS in UNL and reports no issues

[NetworkGroover]

Never used UNL so I couldn't tell you.  I've been perfectly happy with GNS3, so haven't found a need to look elsewhere yet.  Though, I won't be running any VSRX or IOSV Only so much time in the day and right now it's spent on work and learning how to use Ansible to switch my 5-switch leaf/spine environment from L3 ECMP to L2 MLAG and back again.

[calaesha]

UNET lab has a lot of hope, but dear lord they need help. The technology is there, but it's just two Russian guys and they really struggle with documenting their product. It requires a decent level of Linux skill, not too much, but the whole converting VMDK to QCOW2 images can get old quickly. Luckily it only needs to be done once. Their UI is within a web browser, which is AWESOME, but leaves a lot to be desired.

GNS3 has come a long way, and I firmly believe it's the best product for constrained wallets. You'll still need the whole IOU thingy too, but that's not too much of a pain once you figure it out.

Cisco VIRL does not get as much love as it deserves, IMHO. much of this is due to the price point and resource requirements, and I completely understand that point. But here are some killer features of the product:

• can emulate anything x86 based (palo alto, load balancers, JunOS, the list goes on and on http://community.dev-innovate.com/t/3rd-party-vms-in-virl/3369
• incredibly powerful web interface. allows for nifty things like visual trace routes (management loves these demos)
• ability to add latency and packet loss on links. really awesome for real world networks
• has an OpenVPN server built into the service. allows you to VPN into your home server from anywhere and build/test/destroy
• native Python on box. Hell, you can deploy a Python script that will create a topology, light it up, and shut it down after a period of time
• AutoNet configuration builder, a personal favorite. just drag and drop nodes onto network, VIRL will build the basic configurations for you per your spec.

i have justified the cost of VIRL's annual subscription from the last mentioned feature alone. no more spending an hour assigning IP addresses, AAA, routing protocols, etc. drag, drop, deploy.

Node count is a real limitation, but it only counts against Cisco devices (Arista, Juniper, A10, etc. don't count). It's currently at 30, which should be more than enough to tackle real world stuff.

[SimonV]

Just got it running in Unetlab with version 4.13.14M, sweet!

At first I thought something went wrong because I didn't see the Ethernets with a show ip int brief, but turned out to be false alarm and I needed a different command. Glad to have a working switch!

[NetworkGroover]

Just got it running in Unetlab with version 4.13.14M, sweet!

At first I thought something went wrong because I didn't see the Ethernets with a show ip int brief, but turned out to be false alarm and I needed a different command. Glad to have a working switch!

Nice! 

[wintermute000]

I use VIRL if I need it to be 100% correct e.g. mirroring a production change (or I need NXOS/ASAv or the day I need IOS-XR).
The features you noted are indeed a killer. I've been subscribing since before it went version 1.0.
For day to day R&S feature labbing, esp well known stuff like BGP or OSPF etc. then I just use GNS3/IOU.

The only downside is that due to its HW requirements, I can only really run it @ home and work is locked down to the nines (big telco) so I have to jump through multiple hoops to VPN back home, which also involves remembering to turn my lab on before I leave the house. Whereas GNS3 is light enough to just fire up on my laptop on a whim.


Note: with UNL there is nothing stopping you from setting up python/ansible on the UNL VM itself.

[calaesha]

Apologies for derailing the original post, but I just realized that EVPN is supported within VIRL through IOS-XR. Well hawt-diggity-damn. I'll need to test whether or not the vMX solution supports it as well.

[wintermute000]

only control plane, but thats enough!
Protip: need at least 8 cores and 48Gb RAM to run the INE CCIE SP lab properly

[calaesha]

only control plane, but thats enough!
Protip: need at least 8 cores and 48Gb RAM to run the INE CCIE SP lab properly

And that's precisely the moment when God created rack rentals.

[wintermute000]

... or a DL380 G6 in your garage. 12 cores and 64Gb RAM for around 800USD. I've run an entire Vsphere/Vcenter/NSX lab nested (7 nested ESXis + Vcenter + Active Directory) and it didn't skip a beat.
You can probably go one better these days and get Sandy era Xeons, mine are Westmeres. I really, really hate rack rentals, esp. if I'm just renting someone else's VMs... yeeech.

BTW the EVPN in VIRL is L2VPN control plane over MPLS. NOT the EVPN control plane over VXLAN in NX-OS (and all the good stuff like inter VXLAN routing) which unfortunately is more relevant to the bulk of us not working inside SP core networks.