Storage Networks

Scott Shimomura

The Biggest Loser: The Worldwide Data Center SAN Challenge

by Scott Shimomura on ‎01-09-2014 09:24 AM - last edited on ‎02-19-2014 11:37 AM by Community Manager (2,930 Views)

Weight and size matters in the data center.  As we speak, there’s an obesity epidemic taking place in the data center that has the potential to challenge the health and well-being of data centers throughout the world. In the networking world, there’s been a steady increase in weight and rack sizes, leading to unhealthy consequences like decreased reliability, increased complexity, and higher requirements for power and cooling.

This week’s contestant, the Cisco MDS 9710, hails from Tasman Drive. Over the past couple of years, the MDS 9710 has had the opportunity to train under the guidance of the Brocade DCX 8510 with Gen 5 Fibre Channel. The DCX 8510 has been the market leader and industry role model for data center directors since its introduction in 2011. Its efficient and powerful design has raised the bar for reliability, performance, and scalability.

 

 

dcx-weight.png

 The MDS 9710 finally emerged from training and made its data center debut in April, 2013. Curiously in the transition between MDS generations, the MDS 9710 went down the design path of more complexity and fatter is better…

At the weigh-in, the MDS 9710 tipped the scales at a whopping 415 pounds, 40 pounds more and 144 ports less than the previous generation MDS 9513. It also grew 6 inches in depth, now measuring a massive 34 inches deep.

The MDS 9710 was run through a series of challenges to find out whether it is fit for deployment or needs to go back to the drawing board. The Brocade DCX 8510 with Gen 5 Fibre Channel was used as the benchmark for ideal design.

Challenge #1: Weight and Size

Weight and size are indicative of design efficiency and system complexity. In this case, less weight and smaller size equals more efficient use of space and more simplicity.

 

 

dcx-weight2.png

 

Results of Challenge #1:  Cisco MDS 9710 Fails

The MDS 9710 is an unfortunate example of data center obesity at almost twice the weight and nearly 50% greater in size (depth) than the DCX 8510. The MDS 9710 requires special handling with heavy-duty hydraulic lifts capable of handling upward of 500 pounds and racks capable of holding its excessive size.

Challenge #2: System Complexity Challenge

System complexity is a function of the total number of assemblies (fabric cards, power supplies, fan modules, etc).  The overall serial reliability of a system decreases as the assembly count increases. While a system can be made highly available through redundant assemblies, this will lower overall system serial reliability due to the additional number of assemblies.

 

 

dcx-weight3.png

 

 Results of Challenge #2: Cisco MDS 9710 Fails

 By definition, a director-class switch requires redundant hardware. Adding additional redundant hardware assemblies come at the expense of lower system reliability and additional complexity. It’s a tradeoff that has little value based on the incremental improvement in availability.

 Challenge #3: Component Complexity Challenge

 Component complexity is a function of the total number of electrical components (ASICs, integrated circuits, transistors, etc.) on an assembly. Nobody says it better than Cisco in describing the relationship between the number of components and spacing between the components as it relates to reliability and heat dissipation (at 22:25 http://www.youtube.com/watch?v=LkP5EJiPSC8 ), “… that the more components you have on a board, the more failure points that you have. Any good design is going to try and limit the amount of components here and give you a lot of nice green space…”

 

 

dcx-weight4.png

 

 Results of Challenge #3:  Cisco MDS 9710 Fails

Using Cisco’s definition, the MDS 9710 port blade has significantly more components and much less green space than the DCX 8510 port blade. Given the number of heat sinks, it’s also likely drawing more power and generating more heat, both contributors to lower reliability.

Final Verdict

Rather than adopt the training provided by the Brocade DCX 8510 with Gen 5 Fibre Channel, the MDS 9710 turned out to be an overweight, oversized, and over-engineered product.  Cisco designed the new director with the intention to drive new customer benefits, but does it?

 Reliability and availability continue to be the most important attributes of why customers continue to buy Fibre Channel technology. Cisco makes the dubious claim that the MDS 9710 is “the industry’s most reliable storage director”. However, based on the results, it’s clear that the MDS 9710 failed in every challenge. Reliability is proven over time in the data center, not on paper. Even on paper, it’s is obvious that their system has not been designed with high reliability in mind.

 Final Recommendation: Brocade DCX 8510

Clearly, if you are considering new storage infrastructure or replacement of legacy SAN technology, the Brocade DCX 8510 with Gen 5 Fibre Channel should be at the top of your list, especially if reliability and availability are important to you. It is an efficiently designed, elegant, and time-tested director that is purpose-built for data center storage. In addition to hardware design advantages, the DCX 8510 with Fabric Vision software technology will further enhance reliability and availability through advanced monitoring, management, and diagnostics.

Comments
by Community Manager on ‎01-09-2014 09:26 AM

Scott,

So, to summarize, "late, overweight, fat, and expensive".  To paraphrase a gallows humor quip, "So, other than that Mrs. Lincoln, how was the play?"

In my experience, a product with these "qualities" indicates a lack of passion and committment to the market and/or a product team that's lost it's way.

by AJ Casamento on ‎01-09-2014 09:29 AM

Folks,

  A few quick observations here:

  1. That kind of weight and depth also constrains "where" in a rack customers would be willing to place the product. You would not, for instance, want to place such a unit high in a rack that only contained 1U rack optomized servers. Such a placement would make the rack top heavy and be a safety issue.
  2. Most of my customer base does not have unlimited power deployment for a rack. Especially in  so called "brown field" datacenters power density per rack is more in the 10-12KW range. As I understand the base config of the 9710 it would need 3 of the 6 3KW power supplies for a full chassis. Using 75% of the power service to a rack in the first 14U of the rack begs the question of "What do I use the rest of the rack for?" Most of my larger datacenter customers want to use all of the rack footprint and certainly any hosting or cloud provider understands that every rack unit you can't sell is margin you don't make.
  3. On the reliability discussion, unless something has changed drastically the Bellcore mean time between failure predictors is an arithmetic calculation based heavily on component count in the design. There is no way one can torture the data sufficiently for port blades with so high a volume of discrete components can compare to the MTBF of our port blades. Never mind the power and cooling issues (which contribute to component wear and failure).

Just some practical observations from the floor of my datacenter customers.

ciao,

aj

by mangelin on ‎02-04-2014 10:21 AM

 

Evaluator Group recently completed testing of storage networking connectivity between blade servers and solid-state storage, evaluating Fibre Channel (FC) versus Fibre Channel over
Ethernet (FCoE).

 

Within the report Evaluator Group outlines their results, showing that Fibre Channel connectivity is required in order to achieve the full benefits that solid-state storage is able to provide.

 

The tested configuration showed the following results:

 

  • 50% faster response times under load
  • Higher performance with 50% fewer connections
  • 50% less power and cooling
  • Used 20% - 30% less CPU

 

Labels