Cisco UCS (Emulator) – a rant with hope.

Cisco is offering a great simulator for UCS that lets you poke around with VLAN/VSAN and other provisioning and also peek at the hardware management.

I like it a lot – it strongly reminds me of the Sun N1 management solution that (those idiots) didnt offer for the Sun Blade modules when they invented the whole thing. Anyway noone remembers and so lets just not sob about it when it’s already over. Cisco UCS is there and they seem to be serious about it.

Concerning UCS I had been trying to ignore it because

  • I smelled marketing buzz
  • It was strongly oriented towards VMWare which is great for cisco since they’ll sell more blades due to the added overhead
  • cisco killed the infiniband offerings hard to avoid having their own QDR infiniband offerings eat into the slower UCS networking

But I couldn’t completely avoid to drool at the UCS servers which are so nicely fit for heavier-duty virtualisation. The cisco UCS blades look much like the awesome ones Sun offered in the last Berchtolsheim era. Unfortunately none of the UCS systems have yet made it into the used, err affordable market. I’ll leave to the reader whether this is just because their customers are 100% happy or maybe also due to low sales volume.

So how did I finally get around to UCS?

I started writing a Vlan provisioning system in python and wanted to double check I’m not reinventing the wheel. Surprisingly my searches ended up almost empty handed when looking for such tools in perl or python. Good for my project but still an odd thing. I figure in most conpanies the devopsadmins still dream their little dreams but the reality must be far, far, behind.

At my last customer they’re now buying into HP datacenter (aka “runbook”) automation big time (and money) to achieve what the infrastructure administration movement (steve traugott, mark burgess, etc.) had managed just like 10 years ago: couple your whole IT and be able to really deploy stuff from end to end.

I guess that is what will sell cisco UCS above all other things – you need a network vendor to get the integration to work.

You want FC performance and stability, but you only want two cables running out of the box, even with dozens of vlans and all the shortcomings of ethernet in the way.

For that amount of reliability datacenter bridging, Trilll and QinQ, FCoE will be needed, you need to have switches that offer much more than a normal ethernet can do (virtual stacks at the very least). Some people will suggest to flatten the datacenter lan with switch-embedded firewalls and stuff, some others (i.e. cisco) publish patents for running MPLS throughout your whole enterprise core lan.

Suddenly you end up asking yourself if a vendor like HP or IBM can even remotely “unify” all the networking needs of a more flexible network. Let alone the smaller vendors or those without any in-house network knowledge. Re-selling brocade only goes so far.

Reality example:

you’re to have a VM in each of your datacenters and these two should see the same storage and share a few public vlans and a few just between the two of them.

this is supposed to be carried through the routers and yeah, it’s supposed to happen automatically after filling a request with the needed data.

you bought a “dynamic datacenter” from a vendor with almost no inhouse skill, that got you a bunch of blades, a few vmware licenses and a shady management application that can install blades if you click through the wizard.

=> WHO will pay for making this happen automatically?

You. You just saved about 20% on the hardware + software stack over UCS and now it might help the career if you can show off the cool 3rd party tools you bought to try bringing together the whole MESS.

But: it will never work well and essentially you FAIL for wasting a lot of money and time, and your vendor will be laughing their ass off because they can actually sell you MORE services and hardware to make up for their products non-existant flexibility…

And now we finally spin around to UCS. It was almost the only thing in IT-land that is able to be interfaced and create a few thousand vlans on the fly and assign them where needed. That caught my interest, big time.

At the launcher website there’s the XML API documentation and a link to the web ui webstart.

That’s it. Right at the first possible occasion they go ahead and give you all the tools to script, interface and do magic.

Ok, let me admit the Java UI is ugly as hell. Still I’m shellshocked at how cool a fresh design can be. I’m not even saying I’d ever chose UCS myself – I like playing with capirca and my VLAN magic tool and cobbler and a few more OSS tools will get me exactly where I want, which is different and beyond from UCS, and most importantly, it lets me keep my infiniband stuff.

But I really hope and pray that many of the companies with their large datacenters, large pockets, and little foresight will have a visit from a cisco sales rep and, even if they don’t grasp it technically, maybe their guts will make them notice HOW far the differences go between a pile of incompatible crap and bad tools and an elegant “computing system” that comes with the API docs at the login page.

Do something for the environment and save a few billions in consulting hours….


2 thoughts on “Cisco UCS (Emulator) – a rant with hope.

  1. Wow, Very honest and real – Enjoyed reading it… If only the larger enterprises… I work for a fortune 5 company and know exactly what you mean. UCS makes a lot of sense in some ways but they are so used to doing the good ol’vendor crap…

  2. Hi,

    so where should we try to change things?

    Right now, I think the most important change would be having less NDAs so fortune NNN admins could at least openly discuss 😉

    I have been thinking about it for a little while and re-editing this comment a few times hehe. Maybe You can compare the below to what happens in your world.

    The management said “we want automagic to be turned on”.
    -> that way the unix head admins went looking for some tool. various requirements were added later and first abused the tool till it was close to useless, then it needed to be extended to the network/san scope for $$$.

    The didn’t say “turn it on for unix and linux servers using the same tools and you must include our campus lan and san config and vmware or xenserver, as needed”. Ensure it scales to reinstall up to 100 servers in one night if disaster strikes.

    In that case this suddenly reads
    “ah cool, sounds like the should look into NPIV right away, and the need some api-magic for NetAPP and HP (oh, then we can’t just use netapp api, we need SNIA) and vmware should be fine.
    Then around comes a ressource planning for integrating all these single pieces… and one might actually think … how about we use UCS and we got the LAN, SAN, VMWare part covered saving us more than a few man-months?

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s