 Tony Rybczynski - Director of Strategic Enterprise Technologies for Nortel, is the author of an interesting enterprise opportunity blog hosted by TMC.net:
The Hyperconnected Enterprise Blog In his most recent blog entry, Tony suggests reading the fine print of the Cisco Nexus 7000. |
Tony's sampling of the fine print:
 |
High quality right? The Nexus is a new technology for the core data center, with new hardware and a new unproven OS (they call it release 4.0, but this isn’t very convincing). |
|
At 15 Tb/s capacity, isn’t Nexus all about performance? Look again. One slot has a max I/O of 230 Gb/s and with 10 slots per shelf, my math says 2.3 Tbps is the real capacity. And it’s an energy hog in spite of Cisco’s claims to the contrary (based on what?): our high-end switches have 185-320% greater energy efficiency on a per port basis. |
|
Ok, so is it ultra-reliable for data center applications? Wrong again. It takes the Nexus 7000 4-5 seconds to recover from network failures (it doesn’t even support the much touted VSS capability). And because it has no intelligent services virtualization and limited module options, building real data centers has just got more complicated. |
|
So if the Nexus is no Lexus, with questionable quality, performance and reliability, then what exactly is it? It’s touted as a unifying data center platform but doesn’t even support Fibre Channel. Or does it mark the beginning of the end for the Catalyst 6500? |
View Part 2 of Tony's blog entry.
Do YOU concur with Tony's summation of the Cisco Nexus 7000 fine print, that the Nexus is no Lexus?
About Brad Reese on Cisco
Brad Reese is research manager at BradReese.Com, advancing the careers of 1 million certified individuals in the growing Cisco Career Certification Program.
Contact him.
Brad's blogroll
Brad Reese on Cisco archive.
Cisco Subnet
|
|
Nexus 7000 Information
Hey Brad! I wasn't going to deign to comment on one vendors slanted perspective, but since you picked it up I'll give it a whirl. :)
A few comments-
On quality and 4.0 versioning-- We built NX-OS on top of SAN-OS as we've stated several times, I am sure that in researching the fine print the Nortel product manager found that SAN-OS is on major version 3.x It started with 1.x coincidentally and in the past 6 years has moved from 1.x to 3.x. We felt that since the entire core SW architecture was based on SAN-OS 3.x that adding modular and multi-threaded L2 and L3 code would probably warrant the next number iteration. And since the SAN SW components are consistent with 3.x didn't want to run into version numbering overlap. I hope this is rather intuitive.
On capacity-
Each payload slot is 230Gb in and 230Gb Out. I am pretty certain every other company would call that 460Gb, but I've always personally stuck with the real number for per-slot bandwidth. Each supervisor slot, coincidentally has 115Gb too, just FYI. The 18-slot chassis, coming in a couple of quarters is also 230Gb/slot. But more importantly than that the system architecture has already been tested to far more than 230Gb/slot and we know this system has significant headroom. Most other vendors churn their infrastructure 4-5x faster than we do, or offer little to no innovation to the networking space. We just wanted to build a system that could last and do it well.
On the energy efficiency claims- please check the math, I mean seriously... Also be sure to look at actual draw versus provisioned power. This is important, and there is a good lesson here for folks, because you need to provision more power to a device so you can handle PDU/Grid and Power Supply failures. Although the actual draw and heat generated will be markedly less. Someone also show me the math, I only got up to algebra mind ya, that gets a 300% power savings... hmmm... if I took 10w and multiplied it by the savings of 300% I would get.... darn this is hard.
4-5 seconds to recover from network failures?
Not sure where this number came from. An OSPF failure on the box is detected and corrected in less than a second in many cases not dropping a 'Hello' packet. A administrator initiated system shutdown will PRECONVERGE the network rather than forcing RECONVERGENCE (Another Cisco innovation). The Supervisor can fail and may take a couple of seconds to get everything live on the second supervisor but since the Supervisor has no uplinks, no forwarding engine, and no switch fabric there is zero-packet loss when this occurs.
This poses an interesting question- if a supervisor fails and no packets are lost do end users care?
We are taking a different approach to Multi-Chassis EtherChannel with the Nexus family that supports even more x-sectional bandwidth than VSS does.
On services modules I thought we were quite clear- its hard to build a fast enough fully capable firewall in a power budget of 600-800w to utilize enough of the backplane capacity to make it worth the physical integration and common equipment cost-points. So we continue our investment in the Catalyst 6500. Also, I appreciate the validation of our services module strategy from companies that don't have one... thanks!
The Nexus 7000 will support FibreChannel over Ethernet and has the architectural capability to support FC as well although we have not made a decision to do that yet, we instead wanted to preserve our investment in the MDS 9500 SAN Directors, especially given the Intelligent Fabric Applications running there. Amazing this comment would come from Nortel- I searched all over for their SAN director...
Certainly companies that have consistently failed to innovate and deliver in the networking segment, that have married their own R&D capabilities so tightly to the merchant silicon vendors that they have no capability for competitive differentiation and have milked one business after another to prop-up flagging businesses or to build me-too offers have a right to be heard. I would offer that rather than trying to use our creation of a new category of data center infrastructure as an opportunity to try to deposition the Catalyst 6500s position as the preeminent switching platform, especially in the campus, data center access, data center services, metro, and branch please focus inwards- meet with your customers and listen to THEIR problems, and build something innovative. Ship it. Deliver. Then draw a comparison.
dg
Nexus is no Lexus- Part 2
I posted my original blog (the Nexus is not a Lexus) because 90% of what I read on blogs and in the trade press was just a reiteration of the Cisco press release, without any critical examination (for example, the Nexus today is touted as a unifying data center switch, but doesn’t support Fibre Channel).
Thanks Doug for contributing to the discussion.
So what are my new takeaways?
Today, Nexus doesn’t and won’t support Cisco’s VSS, the just announced partial response to Nortel’s innovative MLT technology. We see network reliability within and between data centers (and across campuses) as absolutely critical, with our customers demanding very fast failure recovery using a common end-to-end architecture.
Link to Phil Edholm’s VSS blog at
http://blogs.nortel.com/enterpriseblog/2007/12/06/cisco-vsstoo-little-too-late/
Energy efficiency is not a ‘damn hard’ calculation. 100% better efficiency means twice as many ports per watt. What may be damn harder is for enterprises to find their way through vendor energy efficiency claims. That’s why we believe in third party validation. Take a look at what the Tolly Group measured comparing Cisco with HP with Nortel. Stay tuned for similar third party validation involving the 6500.
Here's the link to the Tolly report:
http://www2.nortel.com/go/solution_assoc.jsp?segId=0&parId=0&catId=0&rend_id=22601&contOid=100183401&prod_id=38820&locale=en-US
Finally, while some may debate the degree of innovation in Nexus, it’s interesting that it came about from internal Cisco R&D, rather than the pattern over many years of innovation by acquisition.
Tony PLEASE!!!!
Tony- get a clue man. You guys are not even in the switching space! You have a switch with 30Gb per slot but you call it a 720Gb backplane? Can you explain that math!!! I tested one for my employer and when you pull your fabric I lose half the bandwidth! Is that a 240Gb or 120Gb design? How do YOU get to 720Gb? Please also upload a network diagram and the math on how you get to 2Tb per your claims as well.
Please explain yourself before casting some stones.
Split-MLT is a hack too. Software forwarding of broadcast and multicast almost brought the control plane to its knees. Not even close to production ready for any enterprise.
Innovation? Nortel? ARE YOU KIDDING ME!!! You acquired Bay, which was a merger of Synoptics and Wellfleet. What innovation have you brought to the market??? I saw some interesting powerpoint though.
Looks like Gourlay owned you... not surprising.
Oh, didn't Tolly advocate Fast Token Ring as being superior to Ethernet? Try getting an analyst that actually analyzes things, not a pay-for-play windbag.
Nortel just got OWND.
HAHHAHAHAHA
Tony - thought you could use some help!
Since you are obviously way behind on technology and current trends, I thought I'd send this link to you to help you understand how you're being viewed after sharing your thoughts...
http://www.urbandictionary.com/define.php?term=ownd
Tony, You should get out of
Tony,
You should get out of the blogging business. Every time you open your mouth Doug sticks his foot in it and you look like an absolute fool in front of the world.
NANOG
btw- I just wanted to invite everyone who is interested in learning more about this, having these questions answered by the key engineers who helped deliver the platform, including folks like Dino Farinacci (lead SW architect for NX-OS and Cisco Fellow) and Venu Venugopal (who led the L3 SW engineering team and also was one of they key SW leaders at Procket). Please come by NANOG's Beer and Gear Monday night 5:30 - 7:45pm at the San Jose Fairmont.
We will have a Nexus 7000 Data Center Switching system there, the key engineering team, some good beer, and great t-shirts and would be happy to answer any questions anyone would have. Fact-based, fun, and informative...
Wow
Tony, apparently getting your ass kicked everyday in the marketplace wasn't enough for you. Nice job on getting it kicked in a public forum as well. It's always a great idea to put up a weak, uneducated argument while represening your company just to have it jammed down your throat by someone who actually knows something. Good luck passing off MLT as virtualization. Let everyone know how that strategy works out for you in a few years. What's that old saying...."doing the same thing over and over and expecting different results is the definition of insanity".
Math and Lies
While not an expert in 'Green' math, I would have thought there was more to it than Watts per Port. If I had a 100mb port, 1Gb port and a 1Tb port, do I calculate it on Watts per port even though each port is doing successively more work?
After all, 1Gb is doing 10 times more work than a 100Mb port and 1Tb is doing 1,000 times more work than a 1Gb port and 10,000 times more work than a 100Mb port. One box may be replacing many boxes and so overall, lower power consumption
Therefore another calculation that should be considered would be Watts per Mb or Gb.
Point is, clearly this is a little more complex than Tony is making out, but of course, that would kill Tony's argument.
Finally, on that last note, Tony seems to spend an inordinate amount of time negatively commenting on competitors on blogs - often caustically. Just as with elections, this is a sign that the candidate is losing.
Whatever the right calculation
Good Point Paul!
You are right in that each successive speed iteration in transport technologies does drive performance up 10x but usually at a 2-3x power consumption. The other factor to consider is the port capabilities-
i.e. Buffering, Security, Encryption, Access Control, Quality of Service, Size of Routing Tables, Size of CAM Tables, MPLS Support, VPLS Support, etc. These are all factor that drive increased numbers of transistors in the silicon architecture, take memory space up, and thus use some amount of power. (30% or more of a board power draw can be these types of memory assemblies.)
So it's hard to compare the types of features and capabilities you would get on say one of our 4-port Linksys 10/100/1000 switches designed for the consumer market to a Nexus 7000 designed for Service Providers and Enterprises. Unfortunately that is exactly the type of comparison some people draw when they don't take the time to understand the requirements for each customer segment and do an apples-apples comparison.
Here is one of my favorite glaring examples of exactly this- http://blogs.nortel.com/enterpriseblog/2008/02/21/merchant-silicon-benefit-or-bane/
I would correct the author in that using a more reputable analyst firm and quoting your sources is usually the best course when throwing around market share statistics- so if you check Dell'Oro who I believe is one of the most highly regarded quantifiable research firms for networking you will find that Cisco is very fortunate to have 75% revenue share and 71% port share in Modular L2/L3 switching. A far cry from the authors claims, that I am certain include low-end consumer equipment if they are even valid.
So as we look at driving efficiencies we need to be cognizant as you point out of the speed and increased workload component, but also of the services component and draw a fair and apples to apples comparison. This is especially important as the network starts supporting lossless transmission of FCoE and helping reduce the overall power draw of the data center by more power than the network consumes.
dg