Arista vs Nexus, which would you rollout?

[NetworkGroover]

Just got a list price quote from arista. VERY EXPENSIVE. We are going to be looking at their other options (were looking at their 7504R platform). It was going to be like ~400K list. lol.

That's list price, and for the best platform the company has to offer - which blows away the 9500.  Instead of just looking at price tags, evaluate on technical merit.  There are reasons why Cisco hates Arista, reasons why they drop pants on pricing when they can't compete technically, reasons why they continue to lose market share in the DC space, etc. - take all of that into consideration.  Do you understand the capabilities of the 7500 or did you just throw up when you saw the price?  Have your rep discuss the technical competitive points versus the 9500, and if you still don't like it, Arista has other platforms as well.

Evaluate on technical merit first - THEN shift if you have to due to budget.

[that1guy15]

I prefer the Facebook wedge as it has all those different color blinky LEDs on the front.

Does anyone else have cool blinky lights? uh huh...

[LynK]

I am not going to sit here and say I know everything about the product line, but I do know that the 7500 series does have significantly more functionality than the 9504 systems.

[deanwebb]

At the end of the day, however, accounting is architecture. If you want to justify actually spending money on hardware vs. free hardware, you've got a lot of homework ahead of you. Cisco knows this, which is why they cut way back on their purchase prices, hoping to make the money back in licensing renewals and professional services engagements. In accounting-speak, this is all part of TCO, total cost of ownership.

If you can put numbers on *all* the costs and show that vendor A is going to be cheaper than vendor C, J, or even B, then you'll have numbers the accountants can understand. Any savings on personnel costs, maintenance, licensing, and the like will be important to identify and include.

Think of it this way: if we spend $400K on gear I can configure myself, that's ultimately cheaper than free gear that requires a $500K 6-month pro serv engagement with the vendor in order to get up and running. If you estimate more vendor staff would be required, say an actual on-site, year-round technical resource, that can easily hit a million, depending upon the vendor. That million may include all-hours tech support and a case manager, but that's still way more than the gear that has a higher cost that you can set up on your own with a minimum of tech calls.

[icecream-guy]

remember my Credo (LoL) one of my old companies stood by this,
buy the cheapest POC you can, and spare no expense implementing it...
If it costs 10K up front and 50K to make it run, that's much better than 50K up front and 10K to make it run.
Take this advice with a grain of salt.  it's probably better to do the opposite.
But the original take makes the bottom line look better upfront.

[wintermute000]

Look honestly, it sounds like you're going shopping before you've ironed out your requirements and design. This is a big no-no in my book. The book (ok, the consulting book, but it works) says: requirements then analyse options then design then FINALLY a BoM that derives from the design.

Do you really you have requirements that are unlikely to be met by any old Tomahawk based (or that equivalent performance e.g. Cisco 9200/9300s) fixed 10/40/100G merchant silicon? I would love to see your requirements and environment if 1 and 2RU pizzabox switches that can handle 48x10G and 4x40G uplinks nonblocking (raw or VXLAN encap) are too slow and you have > 16k MAC addresses per fabric. And off the top of my head I believe that's Trident2 figures i.e. 2014 merchant silicon, not 2017.

In other news, Cisco has finally gotten their ESI (Type-1 and Type-4 EVPN advertisements i.e. active-active multi-homing to the same LAN segment with EVPN without MLAG/vPC) working, and that despite all the white papers and the April 2017 textbook saying no, that EVPN DCI is now available (so EVPN multi-fabric with a EVPN DCI segment in the middle... sounds funny until you read the old 'solution' which was OTV, ugh). Expect formal public announcements in the near future.

[LynK]

Listen Guys,

I know how the proposal process goes. This is still very early in the project phase. I am looking to get numbers/prices for a project, to see what the cost would be so when presented to ownership I can give them a budgetary number, and answers to our solution. This is where I am now. The purpose of this post was not for your assistance with this project, but rather in your professional experience what are your thoughts on Arista's platform. How does it meet XYZ functionality, how is their support, pros/cons of their hardware/features.

Nothing more nothing less.

As far as our infrastructure currently. We are 99% physical server infrastructure, with a collapsed 6509 core (yes... non E). The idea is to provide the owners with an attractive upgrade to infrastructure that is current, and supported. My idea is to do this as cost effectively as possible. I do not want to drop a 400K deployment, and say here ya go. For example, I have requested a quote for a refurbished 9504 chassis with 23XX FIs. This I can get for around 120-140K list, and ~ around 80k, which is a very attractive price.

I am also investigating non chassis based deployments as well.

[deanwebb]

How many of those servers are on port channels so they can pool multiple NICs?

[burnyd]

In other news, Cisco has finally gotten their ESI (Type-1 and Type-4 EVPN advertisements i.e. active-active multi-homing to the same LAN segment with EVPN without MLAG/vPC) working, and that despite all the white papers and the April 2017 textbook saying no, that EVPN DCI is now available (so EVPN multi-fabric with a EVPN DCI segment in the middle... sounds funny until you read the old 'solution' which was OTV, ugh). Expect formal public announcements in the near future.

Read that non biased non fake news link I posted.

[burnyd]

Listen Guys,

I know how the proposal process goes. This is still very early in the project phase. I am looking to get numbers/prices for a project, to see what the cost would be so when presented to ownership I can give them a budgetary number, and answers to our solution. This is where I am now. The purpose of this post was not for your assistance with this project, but rather in your professional experience what are your thoughts on Arista's platform. How does it meet XYZ functionality, how is their support, pros/cons of their hardware/features.

Nothing more nothing less.

As far as our infrastructure currently. We are 99% physical server infrastructure, with a collapsed 6509 core (yes... non E). The idea is to provide the owners with an attractive upgrade to infrastructure that is current, and supported. My idea is to do this as cost effectively as possible. I do not want to drop a 400K deployment, and say here ya go. For example, I have requested a quote for a refurbished 9504 chassis with 23XX FIs. This I can get for around 120-140K list, and ~ around 80k, which is a very attractive price.

I am also investigating non chassis based deployments as well.

You 100% do not need chassis now a days.  Vendors have priced themselves out of chassis with the merchant silicon lanes.  So since we are talking about Tomahawk in its first can handle 132 ports of 10GB links in its second generation that will be out will handle a lot more.  Keep in mind this is a cost effective box thats in 1RU form factor. Breakout cables are not pretty but they are great for the price.

[wintermute000]

Hi burnyd - yes I read it - its great but if I go on vendor doco and current design guides etc and they say no go I can't very well assume its all great, esp as the testing may well have been on pre-release HW/SW revisions that I either can't get hold of or are not yet supported. I'm getting clarifications from various vendor SEs now and that's the gist of it - expect release second half FY17 and all the guides to be updated etc.

re: Lynk - point noted so I'll back off on the design / process questions and merely address your points directly. re: 9500 with FEX, two things
- what's wrong with a 9300 + FEX if you want to save money? Not to mention the new 9300Ex is now out. Does FEX. Faster and newer. Again I doubt you'll run into any capacity issues with 48x10/25 + 6x40Gb nonblocking.... if you're fine with a 6500 right now....

http://www.cisco.com/c/en/us/products/collateral/switches/nexus-9000-series-switches/datasheet-c78-736651.html

- you are aware that FEX's don't locally switch, even port to port on the same FEX has to hairpin through the uplink? (ristau mentioned something re: local switching coming in for F3 modules in 7.x for 7K but I'm not sure of the status on 9K).

- If you still want to do FEX, for the love of god do not go enhanced VPC. Its marketing designed to make your life miserable during upgrades. KISS
https://rednectar.net/2012/08/30/why-i-wouldnt-bother-with-enhanced-vpc/

I have seen customers go down the 9K + FEX route simply due to price + inertia (CBF with leaf-spine and doing new things, just drop in newer faster iron, happy days).


RE: Arista in general, read Arista Warrior, marvel in what you can do, and remember that book is what 4-5 years old now. If you're not heavily into linux/scripting it may not blow you away. NX-OS has caught up a lot though a lot of where it falls down is at a qualitative level, not a tickbox e.g. yes it has RESTAPI but its flakier than Arista (how to quantify that though without a formal test regime where you go through every feature you could conceivably want to manipulate, I dunno). MLAG has less gotchas/caveats than vPC e.g. can peer routing through it.

The second point re: Arista I'd make is that Cloudvision (esp now that telemetry is out) is amazing. Its by far the slickest turn-key vendor automation solution on the market for DC fabrics. The telemetry is mind blowing - rewing the exact state of the network XYZ seconds ago and check the exact CAM or BGP tables at that state in time.... rollback config network wide.... the list goes on, buyrnd/aspiring can probably give you an even better pitch. If you don't believe me, go bake it off against say DCMN and come back with your findings...

[LynK]

We will look into the 9300 series, thank you for that recommendation. I know all about the enhanced VPC garbage. Trust me KISS is my motto. I am aware that they do not switch locally. They are EXTENSIONS of the fabric.

The primary reason we are looking into a chassis is because we are most likely going to only implement one core in each of our buildings. This allows the FEX design to be much simpler, but also getting the features of a chassis (modular, and easy hardware upgrades) , and also supervisor features (ISSU/etc.)

[wintermute000]

Well unfortunately the 9300EXs are all 1RU not modular.

If you want one monolithic thing with ISSU/dual sup then you're back to the big bad chassis switches.


I can't see though why you can't just have two 9300EXs with single homed FEXs offering vPC in lieu of a big 9500 with dual sup. One is more survivable (and cheaper) than the other?

[burnyd]

ISSU  is the lols

[icecream-guy]

ISSU  is the lols

ISSU with single processor devices