Best CPU options for CML/EVE-NG

[dizzypa]

Let me start by saying I'm studying for CCIE-SP (I already have the EI), and the IOS XRv9000 image is 4 vCPUs/20GB RAM each and I will need to run 20+ of them...

I've been "torn" between the following options that I've found and right now am really wondering if I'll notice any difference from the extra QPI link on the E7 vs E5, or if the memory bandwidth with the jordan creek 2 SMI stuff will benefit me, also the 1st gen XeonSP as a wild card to stay lower in power draw. CML would be running bare metal on any of these options (powered up only as needed) as I have a DL380G8 I used for my EI studies that I will use as my 24/7 for proxmox/random other stuff.

1) HP DL580G9/Dell R930 - 4x8890v4, 1TB RAM (32x32GB so I fully populate every channel with the SMI splitting stuff)
2) HP DL560G9/Dell R830 - 4x4660v4, 512GB RAM (could go 1TB if I had to later on)
3) HP DL380G10/Dell R740 - 2xGold 6152, 384GB (12x32GB, could go to 24x32GB 768GB if I need the RAM in the future)

Any input on HP vs Dell welcome. I know that Dell updates are easier to get for sure, but anything other than that would be good to know.

Thanks!

 

[scline]

I would not worry too much about CPU speed, being a network lab the more RAM the better imo for such a workloads and there is little traffic to process.

Good number of cores and tons of ram is what I would target, once the devices boot they should use little processing.

Personally I like dell for there ease of firmware, but warning. The 900/800 series are going to absolutely chug power. If your using this at home expect a high power bill and associated heat/noise that comes with it.

 

[wesm63]

My previous company built a CML Enterprise lab and it was a 3 node cluster with each node being a UCS C220 M5 having 88 cores and 384GB ram each but that was built to handle 200+ CSR1ks.

Based on the CML resource calculator, you need 40 physical cores and 320GB ram for 20x XRv9k devices. I think your third option would suffice just fine. Cisco is always going to recommend the newest comprable UCS box but I know CML 2.x runs prefectly fine on UCS C220 M5's. (We never built the 200 CSR1ks but were able to completely replicate one section of our network that was extremely complex with around ~70 devices)

No real EVENG experience, but I would image it's going to require the same/similar amount of resources.

 

[dizzypa]

Thanks for the info. I guess this leads to two follow up questions:

1) What is the power draw for a 900/800 dell or equivalent hp server. I have no issues with the cost, just need to make sure I have enough circuit in the basement to run one (iDRAC/iLo for turning it on/off as needed of course). And if I go four socket, what's the primary difference I'd notice between 4600v4 and 8800v4 processors assuming equivalent core counts?

2) The pricing on the 1st gen xeon scalable is not bad on midline gold processors, but the platinums are still pricey as all get out. I can obviously get a much higher raw core count on a 4S server, but is the perfromance per core/virtualization enhancements in the 1st gen Xeon scalable enough to overcome that, or is there a ratio I should work towards?