Countering FUD, IBM's "PureSystem" and IDG

Hi Guys

Update:
I have been provided further info for this, later in the article you will see where I rip apart IBM for claiming that the latency on UCS will be high, http://www.definethecloud.net/inter-fabric-traffic-in-ucs shows that the latency is as low as 6.2Us. Pretty damn low for blade to blade traffic


Update 2: Latency is actually lower than that now http://www.mseanmcgee.com/2011/07/ucs-2-0-cisco-stacks-the-deck-in-las-vegas/ shows that the latency has actually been lowered to 2.0us

 
Not sure if many of you have had the chance to read the "Totally unbiased" Pure Systems vs Cisco UCS research available at:

http://demo.idg.com.au/cw/event/newtechnology/PureSystemsvsCiscoUCS.pdf

(Note, deliberately not linked as I don't want this terrible, totally biased and factually incorrect article from being indexed by google)

It's damn obvious IBM are not even pretending they didn't support this and in actual fact it SAYS it is sponsored by IBM, PAID for, on page 1!!!

"

This document was developed with IBM funding. Although the document may utilize publicly

available material from various vendors, including IBM, it does not necessarily reflect the positions

of such vendors on the issues addressed in this document.

"

So with all that out of the way... let's begin the shredding!

I have quite a bit to say about some of the claims they make below. I guess my main point would be that imitation is the sincerest form of flattery and in this case looks like IBM (also known as Irrelevant-Business-Machines) have really tried there best to provide that same kind of lovely platform that Cisco UCS provides, without success.


Time to rip this to shreds
“The Cisco UCS solution only offers converged Fibre Channel over Ethernet (FCoE) within the Cisco UCS 5100  chassis and not native Fibre Channel. Cisco UCS Manager does enable Fibre Channel over Ethernet in the UCS internal fabric and preserves traditional Ethernet and Fibre Channel connectivity to LAN and SAN environments north of the Fabric Interconnect. However, there is no true Fibre Channel connectivity south of the Fiber Interconnect and there is no true Fibre Channel connectivity within the UCS Blade system chassis. If a customer is currently running native Fibre Channel for SAN connectivity from individual rack mount or blade servers they will need to migrate to the converged 10 GB FCoE (Fibre Channel over Ethernet) Cisco fabric. However, for those customers doing net new installations, this lack of native Fibre Channel connectivity may not pose a problem
“

= Total bullsh*t, you can plug FC directly into the fabric interconnect, not a problem at all. You can even zone straight on the Fabric Interconnect now.

“If a customer is currently running native Fibre Channel for SAN connectivity from individual rack mount or blade Servers”
Unclear, what do they mean, do they mean you have to take ANY existing servers you have and get them to run as FCoE? Of course you don’t, your rack servers can stay the same. If they mean a C-Series Cisco Rack Mount server, then yes you have to buy a CNA for it, but a CNA Means Ethernet and FC over a single bit of wire, which means: Reduced Cabling, reduced complexitity,  reduced port count (2 instead of 4) and reduced number of HBA’s/Adapters required, reduced PCI bandwidth usage too. I just love how they turn an ADVANTAGE into a disadvantage in this clearly “unbiased” comparison.



“However, there is no true Fibre Channel connectivity south of the Fiber Interconnect and there is no true Fibre Channel connectivity within the UCS Blade system chassis.”

Debatable, the CNA shows the OS an FC HBA and a ethernet adapter to the actual OS, the traffic is totally separate and implements Pause frames so I really can’t see this argument holding water. FC Is FC is FC, how its delivered inside the backplane of the chassis is totally irrelevant.

“Customers should be aware that there is no native Fibre Channel connectivity available within the Cisco UCS 5108 chassis or within the rack that contains the chassis. However, there is Native Fibre Channel connectivity that is available on the north side of the Top of Rack Fabric Interconnect switch if an appropriate expansion module is purchased.
“

That’s hilarious, they just took what they said IN THERE OWN DAMN ARTICLE and then corrected it later, proof reading guys, learn to use it. Especially if your going to lie so blatantly.

With Cisco UCS for one blade to communicate with any other blade in the same chassis or another blade in another chassis in the same physical rack, that traffic has to be routed from the first blade up to the Top of Rack FI switch and then from that switch, it is routed to the second blade. In this manner, the Cisco UCS architecture introduces additional latency during blade-to -blade and chassis- to -chassis messaging.
True, chassis to chasis MUST travel up to the fabric interconnect as must blade to blade, but it’s such a large amount of bandwidth and so low latency I cannot believe that IBM would have lower latency on there switches, I would bet my left nut that Cisco’s latency even when you have to go up to the FI then back down again is lower, and considering ALL PORTS ARE LINE RATE, I can’t see that argument holding much water.

Next it mentions all the great profiling stuff Cisco does and VN-Link

IBM PureSystems provide the same level of profiling capabilities in the current release with significant enhancements cited in their product road map
Oh so you have VN-Link? No? Oh well then guess what you DON’T have the same level of profiling capabilities and no subtle references to vaporware “future product releases’ will make that any different

Cisco UCS Manager is a device or element management application that is only available from Cisco with the purchase of a Cisco UCS Fabric Interconnect. UCS Manager handles hardware provisioning, configuration, and management but only for UCS certified components such as Cisco B series blades and Cisco C series rack mount servers. UCS manages these servers as stateless devices and uses XML to configure these stateless devices using UCS specific Service Profiles.
So in order to get UCS Manager.. you have to buy UCS…. Because the fabric interconnects make up part of the UCS< OH MY GOD GUYS WHAT AN INSIGHT! I WISH I HAD SOME HARVARD RESEARCHES TO EXPLAIN SUCH SIMPLE CONCEPTS TO ME TOO! This is creative writing where they try to make it sound like you have to buy a fabric interconnect separately, the FI is part of the UCS system, so DUH, it comes with UCS, so DUH, you get UCS manager, but by wording it the way they have they make it seem like its something additional to be purchased, so much for truth and honesty in journalism huh?

Cloud readiness
is the system cloud ready on initial power up?

HA HA HA HA HA HA HA…. That’s… I mean that doesn’t even MEAN anything guys, come on!

Does the system arrive at the customer site fully configured and ready to do useful work?
Oh yeah sure guys, it just rocks up at your doorstop with everything installed and customized and ready to go.. idiots


Now we get to my favorite bit, the comparisons:

Security;
“BM has built in system security at the chassis level IBM Pure Systems are designed to boot in a security integrated and optimized configuration
.
Cisco
provides some level of access control and here again relies on third parties to
provide the needed functionality integrated through XML APIs
. “

WRONG, RBAC is BUILT INTO UCS and works perfectly, I love there word smith work here too, "IBM Has built in system security at the chassis level and are designed to boot in a security integrated and optimized configuration" Yeah guys the sales wank doesn't impress me, technical solutions do. Your sentence means nothing without technical context.


Shipping: One price, one part number, one shipping container
–
it will be hard for Cisco to catch up with IBM PureSystems in this area.
Oh, so I can’t specify what I want to order? Amount of CPU? Amount of RAM? Prettttttyyyy sure if I start specifying stuff like that it’s gonna need to be a separate part number


Network PureSystems in chassis (East/West) messaging is superior in heavy workload
situations to Cisco’s latency prone North South oriented architecture. Cisco is market share leader in networking and Cisco UCS clearly benefits from this expertise. IBM has made some recent acquisitions but has a way to go to reach parity with Cisco whose principal business is Networking

So they make a claim that its latency prone, despite the fact that Nexus Switches (go look this up) are exceptionally low latency, and in actual fact the Nexus 3000 Switch is used in high frequency trading because it has the VERIFIED, WORLD RECORD LOWEST LATENCY OF ANY SWITCH OUT THERE, _then_ proceed to say “Cisco have the market share leader in networking and clearly benefit from this expertise and say that IBM still has a way to go to reach parity but STILL rates IBM higher?????!?!?!

Update:
I have been provided further info for this, later in the article you will see where I rip apart IBM for claiming that the latency on UCS will be high, http://www.definethecloud.net/inter-fabric-traffic-in-ucs shows that the latency is as low as 6.2Us. Pretty damn low for blade to blade traffic



All in all I’d say thanks IDG for the great toilet paper!