Thursday, May 20, 2010

Networks Go Green and Save You Money

Servers get most of the glory when it comes to energy management, but networking gear is about to catch up.
Over the past year, network equipment vendors have began to emphasize energy efficiency features, something that was never a top priority before, says Dale Cosgro, a product manager in Hewlett-Packard Co.'s ProCurve network products organization.
Networking infrastructure isn't in the same class as servers or storage in terms of overall power consumption -- there are far more servers than switches -- but networking can account for up to 15% of the total power budget. And unlike servers, which have sophisticated power management controls, networking equipment must always be on and ready to accept traffic.
Data center energy stats
How much power does data center gear consume?
·         Cisco Catalyst 6500 series switch, fully populated: 2kW/rack to 3kW/rack
·         Cisco Nexus 7000 series switch, fully configured: 13kW/rack
·         Fully loaded rack of servers, average load: 4kW/rack
Sources: Cisco, HP Critical Facilities Services
Also, networking power use at the rack level is significant. A Cisco Catalyst 6500 series switch consumes as much as 2kW to 3kW per 42u-high rack. Cisco Systems' largest enterprise-class switches, the Nexus 7000 series, can consume as much as 13kW per rack, according to Rob Aldrich, an architect in Cisco's advanced services group. A 13kW cabinet generates more heat than many server racks -- enough that it requires careful attention to cooling.
By way of comparison, most data centers top out at between 8kW and 10 kW for server racks, says Rakesh Kumar, analyst at Gartner Inc. The average cabinet today consumes about 4 kW, says Peter Gross, vice president and general manager of HP Critical Facilities Services.
Vendors have already adopted some energy-related features, such as high-efficiency power supplies and variable-speed cooling fans. But with switches, there's a limit to what can be done in the area of power management today. Most idle switches still consume 40% to 60% of maximum operating power. Anything less than 40% compromises performance, says Aldrich. "Unless users want to accept latency, you have to have the power," he adds.
But huge improvements are coming, says Cosgro. 
More efficient technology
Technology improvements that favor energy efficiency are gradually evolving in several areas. "As new generations of products hit the market, more of these kinds of features will be implemented," says Cosgro. Some examples:
·         More modular application-specific integrated circuit (ASIC) designs that allow switches to turn off components not in use, from LED panel lights to tables in memory. 
·         General advances in silicon technology will minimize current leakage and gradually boost energy efficiency with each new generation of chips. Eventually, says Cosgro, "we should be able to get networking equipment that uses 100 watts today down to 10 watts."
·         The development of more efficient software that consumes fewer CPU cycles -- and less energy.
·         Equipment designs that run at higher operating temperatures to reduce cooling costs.
For example, Cosgro claims that HP's current ProCurve equipment can run safely at temperatures up to 130 degrees -- higher than the specifications for most other data center equipment. "That's driven by requirements of IT managers who want to run data centers at higher temperatures," he says.
Higher operating temperatures may work in a single-vendor wiring closet, but network equipment vendors will need to do a better job of testing in mixed environments before temperatures approaching 130 degrees can be sustained -- especially within racks in the data center. "No one knows how networking and other types of equipment will react when sitting next to servers that displace more BTUs," says Drue Reeves, an analyst at Burton Group.
Today, each vendor tests with only its own equipment.
Another focus: improvements in power monitoring, and management with more granular controls. Real-time power and temperature monitoring is key to any data center and is essential for managing growth. "If something is not right, you want to know about it before a catastrophe happens," says Rockwell Bonecutter, global lead of Accenture's green IT practice.
Management software could be configured to identify specific network equipment, such as voice-over-IP phones, by using the Link Layer Discovery Protocol. The software could then automatically shut off Power-over-Ethernet current for those VoIP handsets at a specific time of day or when the associated PC on each desktop is turned off at day's end.
Another example: Edge switches are typically connected to two routers for redundancy during the daytime, but the network could be configured such that one router goes into a low-power sleep mode at night. The second router would "wake up" only if and when it was needed.
These types of applications represent "a huge opportunity for savings," says Cosgro.
·         Better specs and standards 
Emerging standards could soon help save energy during periods when networks sit unused and will help IT compare the relative efficiency of competing products.
·         Energy Efficient Ethernet
The emerging IEEE P802.3az Energy Efficient Ethernet (EEE) draft standard may offer the biggest bang for the buck by cutting power consumption for network equipment when utilization is low. Today Ethernet devices continuously transmit power between devices, even when network traffic is at a standstill. Equipment supporting the EEE standard will send a pulse periodically but stay quiet the rest of the time, cutting power use by more than 90% during idle periods.
In a large network, that's "a whole lot of energy" that could be saved, Cosgro says.
The standard will allow "downshifting" in other modes of operation as well. In a 10Gbit switch, for example, individual ports that are supporting only a 1Gbit load will be able to drop power down from 10Gbit/sec. to what's required to support a 1Gbit/sec. configuration, saving energy until activity picks up again.
Products built to support the EEE standard should start appearing by 2011, says Aldrich.
·         Virtual NICs
Another emerging technology, the PCI-SIG's Multi-Root I/O Virtualization, provides servers within a rack with access to a shared pool of network interface cards. This happens via a high-speed PCI Express bus -- essentially extending the PCIe bus outside of the server. "Instead of a NIC in every server, you'll have access to a bank of NICs in a rack, and you can assign portions of the bandwidth of one of those NICs to a server," probably using tools provided by the server vendor, says Burton Group's Reeves.
Energy savings will come from increased utilization of the network -- achieved by splitting up the bandwidth in each "virtual NIC" -- and the need for fewer NICs and switch ports, he says. He expects to see standards-compliant products perhaps as early as 2012.
·         Energy efficiency ratings
Finally, standardized measurements of energy efficiency have started to appear on some networking equipment. Juniper Networks, for example, includes an Energy Consumption Rating (ECR) on the data sheets for some of its products. ECR is a draft specification, created by the ECR Initiative consortium. Lawrence Livermore Labs, Ixia and Juniper developed the specification, which measures performance per energy unit for networking and telecommunications equipment.
Both Cisco and Juniper are backing the Alliance for Telecommunications Industry Standards' Telecommunications Energy Efficiency Rating (TEER) specification, which ATIS introduced last year.
However, neither specification has been universally accepted. Juniper supports ECR but doesn't include the rating on all of its products' data sheets. Cosgro says HP hasn't included either of the  energy-efficiency standards in its data sheets because users don't understand the metrics. "What they care about is the number of watts used by the solution," he says.
Another strike against the specifications, Cosgro says, is that they lack a detailed, open rating methodology. That means vendors can choose rating firms with methodologies that best suit their needs.
That kind of truly open specification isn't likely to appear until later this year, when the EPA's Energy Star rating for large networking equipment should be released. (The EPA developed an Energy Star Small Network Equipment specification last December, but most enterprise networking equipment will be covered by the upcoming Large Network Equipment specification.)
According to an EPA spokesperson, the agency plans to announce a draft specification for data centers in June 2010, and in late 2010 for data center networking equipment. The new specification, which will cover components such as power supplies and internal chips, Energy Efficient Ethernet and overall power consumption for a device, will be "the key energy efficiency standard" going forward, Cosgro says.
The easiest way for network administrators to increase energy efficiency today is to buy new equipment, but that's likely to be a gradual process because administrators must weigh the age of equipment against potential efficiency gains. While a 15% cut in energy costs adds up when spread across thousands of servers, on a few racks of switches the total savings are much smaller.
Even for a single rack, the cost per kilowatt usually won't support an upgrade. "Most people will not save enough energy in the short run to justify replacing their equipment," warns Burton Group's Reeves. "Stay on your regular life cycle."

1 comment:

  1. According to Cisco the Nexus 7010 can draw 8,904 watts per system, and you can fit two in a rack. By my math this is 17,808 watts per rack.

    Cisco pulls 139 watts per wirespeed 10Gb Ethernet port. This is amongst the highest and least efficient in the networking segment.

    By contrast the recently introduced Arista 7500 is the lowest power draw at less than 10w nominal, and 13w maximum power draw per wirespeed 10GbE interface.