Showing posts with label data center. Show all posts
Showing posts with label data center. Show all posts

Saturday, 30 August 2014

Data centers are the new polluters

IT managers may be too cautious about managing power, and businesses are unwilling to invest in efficiency, study finds

U.S. data centers are using more electricity than they need. It takes 34 power plants, each capable of generating 500 megawatts (MW) of electricity, to power all the data centers in operation today. By 2020, the nation will need another 17 similarly sized power plants to meet projected data center energy demands as economic activity becomes increasingly digital.

Increased electrical generation from fossil fuels means release of more carbon emissions. But this added pollution doesn't have to be, according to a new report on data center energy efficiency from the National Resources Defense Council (NRDC), an environmental action organization.
Whatever happened to Green IT?

In term of national energy, data centers in total used 91 billion (kilowatts) kWh in 2013, and by 2020, will be using 139 billion kWh, a 53% increase.
NRDC Data center efficiency chartThis chart shows the estimated cost and power usage by U.S. data centers in 2013 and 2020, and the number of power plants needed to support this increased demand. The last column shows carbon dioxide (CO2) emissions in millions of metric tons. (Source: NRDC)

The report argues that improved energy efficiency practices by data centers could cut energy waste by at least 40%. The problems hindering efficiency include comatose or ghost servers, which use power but don't run any workloads; overprovisioned IT resources; lack of virtualization; and procurement models that don't address energy efficiency. The typical computer server operates at no more than 12% to 18% of capacity, and as many as 30% of servers are comatose, the report states.

The paper tallies up the consequences of inattention and neglect on a national scale. It was assembled and reviewed with help from organizations including Microsoft, Google, Dell, Intel, The Green Grid, Uptime Institute and Facebook, which made "technical and substantial contributions."

The NRDC makes a sharp distinction between large data centers run by large cloud providers, which account for about 5% of the total data center energy usage. Throughout the data center industry, there are "numerous shining examples of ultra-efficient data centers," the study notes. These aren't the problem. It's the thousands of other mainstream business and government data centers, and small, corporate or multi-tenant operations, that are the problem, the paper argues.

The efficiency accomplishments of the big cloud providers, "could lead to the perception that the problem is largely solved," said Pierre Delforge, director of the NRDC's high-tech sector on energy efficiency, but it doesn't fit the reality of most data centers.

Data centers are "one of the few large industrial electricity uses which are growing," Delforge said, and they are a key factor in creating demand for new power plants in some regions.

Businesses that move to co-location, multi-tenant data center facilities don't necessarily make efficiency gains. Customers may be charged on space-based pricing, paying by the rack or square footage, with a limit on how much power they can use before additional charges kick in. But this model offers little incentive to operate equipment as efficiently as possible.

In total, the report says U.S. data centers used 91 billion kilowatt-hours of electricity last year, "enough to power all of New York City's households twice over and growing." By 2020, annual data center energy consumption is expected to reach 140 billion kilowatt hours.

If companies used data center best practices, the report states, the economic benefits would be substantial. A 40% reduction in energy use, which the report says is only half of the technically possible reduction, would equal $3.8 billion in savings for businesses.

The report also finds that energy efficiency progress is slowing. Once the obvious efficiency projects, such as isolating hot and cold aisles, are accomplished, addition investment in energy efficiency becomes harder to justify because of cost or a perception that they may increase risk. IT managers are "extremely cautious," about implementing aggressive energy management because it could introduce more risk to uptime, the report notes.

There are a number of measurements used to determine the efficiency of data centers, and the report recommends development of tools for determining CPU utilization, average server utilization, and average data center utilization. It says that "broad adoption of these simple utilization metrics across the data center industry would provide visibility on the IT efficiency of data centers, thereby creating market incentives for operators to optimize the utilization of their IT assets. "

The NRDC isn't the first to look at this issue. In 2007, the U.S. Environmental Protection Agency, working with a broad range of data center operators and industry groups, released a report on data center power usage that found that the energy use of the nation's servers and data centers in 2006 "is estimated to be more than double the electricity that was consumed for this purpose in 2000." It called for energy efficiency improvements.

Best CCNA Training and CCNA Certification and more Cisco exams log in to Certkingdom.com


Wednesday, 23 October 2013

Juniper targets key new switch directly at Cisco Nexus 6000

Based on Broadcom Trident II, QFX5100 will support dense 10/40G and Virtual Chassis for smaller fabric alternative to QFabric

Juniper Networks is expected to soon announce a new switch for top-of-rack applications that supports Broadcom's Trident II silicon for dense 10/40G Ethernet capabilities and competes directly with Cisco's Nexus 6000.

Juniper is expected to tout throughput, latency, power consumption and table entry benefits of the QFX5100 over the Nexus 6001, sources say.

Specifically, the QFX5100 switch is said by sources to include 48x1/10G + 6x40G, 96x10G + 8x40G, and 24x40G with two expansion slots for 4x40G module variations. QFX5100 is also said to have latency improvements over previous generation QFX switches, the QFX 3500 and 3600, which average sub-microsecond latencies.

[DATA CENTER DIRECTIONS: Juniper switching boss talks technology challenges, Cisco Nexus 6000]

Support for Broadcom’s new Trident II silicon, which many in the industry – including Cisco’s Insieme spin-in, Dell and Arista Networks – are building new switches on, means QFX5100 will be optimized for 10/40G and have inherent support for the VXLAN specification for VLAN scaling. Co-authored by Broadcom, VXLAN is intended to scale VLANs from 4,094 to 16 million to accommodate the exploding number of virtual machines in the virtualized data center.

Broadcom’s Trident II chip is designed to support up to 32 40G Ethernet ports and 100+ 10G ports. Ports on the QFX5100 can be configured and channelized to support up to 32x40G or 104x10G, source say.

And as expected, QFX5100 will support Virtual Chassis capabilities via Junos release 13.2X50. Up to 10 member switches can be configured into a Virtual Chassis and managed as a single switch, with increased fault tolerance and high-availability, and a flatter Layer 2 topology designed to minimize or eliminate the need for Spanning Tree and other protocols.

The capability may also allow users to configure smaller fabric “pods” without the need for a QFabric Interconnect device. Indeed, Virtual Chassis will also work on existing QFX 3500 and 3600 switches with the new Junos release but only if the switches are in standalone mode – not as nodes in a QFabric.

Sources say the Virtual Chassis capability will usher in a new Virtual Chassis Fabric (VCF) architecture from Juniper that allows a 20-node mix of QFX5100, 3500 and 3600s, and Juniper EX4300 switches to form a data center fabric without a QFabric Interconnect. As such, VCF is a fabric alternative to QFabric, they say.

The Virtual Chassis capability was expected. The pods VCF produces could be interconnected for scale with Juniper’s new EX9200 switch, Juniper Senior Vice President Jonathan Davidson said last spring.

The 48x1/10G QFX5100 will be available this quarter. The other variations will be available in the first quarter of 2014. Virtual Chassis will also be available in the first quarter of 2014.

VXLAN gateway and Cloudstack integration will be available later in 2014, sources say.

Juniper declined comment.

Best CCNA Training and CCNA Certification and more Cisco exams log in to Certkingdom.com

Monday, 27 May 2013

After Arista and before Cisco/Insieme, Huawei hikes 100G bar again

CloudEngine 12816 supports 192 100G ports, tops last year's Interop intro

Huawei this week unveiled a high-end upgrade to its CloudEngine core switch, which debuted here at Interop a year ago.

The CE12816 is a 16-slot version of the Cloud Engine switch, capable of supporting up to 192 100G Ethernet ports. The previous high-end of the Cloud Engine line was the 12-slot 12812, supporting 96 100G ports.

The Huawei CloudEngine series switches comprise the CE12800 core switches, as well as CE6800 and CE5800 top-of-rack switches.

Bullet How Facebook aims to reinvent hardware
Bullet Interop: the quiz
Bullet Interop planning guide 2013
Bullet Wicked cool things to do in Las Vegas – after the tradeshow
Bullet SDN and InteropNet: Cranking up and running world's biggest temporary network

Huawei's 192-port 100G switch comes a week after Arista Networks rolled out a 96-port 100G switch in an 11 RU form factor. Some viewed Arista's introduction of the 7500E as a prelude to the Cisco/Insieme Networks programmable switches expected later this year.

[ BACKING OFF: Cisco nemesis Huawei retreating from the U.S. ]

The CE12816 provides 64Tbps of capacity. In addition to 192 100G ports, the switch supports 384x40G, or 1,536x10G Ethernet line-speed ports, Huawei says.

The CloudEngine series provides bandwidth of up to 2Tbps per slot, and is expandable in the future with upgradeable fabrics and linecards to a total system switching capacity of up to 64Tbps. The switch line is designed to accommodate data center lifecycle targets of up to 10 years, Huawei says.

The CloudEngine series incorporates Huawei's Cluster Switch System (CSS) feature to virtualize multiple switches into one logical switch, as well as a Virtual System (VS) feature to virtualize one switch into multiple independent logical devices. Together, CSS and VS turns the network into a resource pool, allowing network resources to be allocated on demand, Huawei says.

The CloudEngine series also allows network adminstrators to build large-scale Layer 2 networks with over 500 nodes based on the IETF's TRILL specification for lossless, multipath Ethernet, Huawei says.

Also at Interop, Huawei plans to articulate its commitment to the enterprise market with the "ICT Nation" initiative. ICT Nation will showcase Huawei and partner products to address enterprise challenges such as bring your own device (BYOD), cloud data center, enterprise LTE and software-defined networking (SDN).

Huawei established its Enterprise business group in 2011. It employs 18,000 worldwide and 80 in Silicon Valley, the seat of its U.S. enterprise operations. Huawei Enterprise did $1.9 billion in revenue in 2012.

Huawei executives recently disclosed plans to retreat from the U.S. market in telecom, but the company's enterprise business is not affected by that. These executives, however, downgraded Huawei Enterprise's revenue goals for 2017.

Huawei partners HVE Connexion, Netfast, Primasense, Softnet Solutions and Zadara Storage will participate in the ICT Nation demonstrations at Interop.


Best CCIE Training and CCIE Exams and more Cisco exams log in to Certkingdom.com