Green Data Centers
From Anita Borg Institute Wiki
GREEN DATA CENTERS Location: Torreys Peak III
Panelists: Aglaia Kong (Symantec), Allyson Klein (Intel), Jane Snowdon (IBM), Brinnan L. Taylor (NetApp), Rachel Zhu (NetApp) We are seeing explosive information growth. The demands for more storage, more servers, faster network interconnects, data replication and data protection are forcing the data center to grow larger and more complex. At the same time, energy prices are rising at an alarming rate, and physical space cannot keep up with business growth needs. Developing an energy-efficient, sustainable data center that still maintains high performance is vital to the future of data center development and management. This panel will examine power and space considerations, costs and the latest tools and trends available.
Going to discuss efforts going on between companies to make data centers more green. Symantec, Intel, IBM.
Aglaia, chief architect for storage & availability management group at Symantec Jane, projects targeting small and medium business, looking at the industry impacts of climate change Allyson , data center efficiency Brinnan, now in sales but green data has been huge part of career Rachel, reference architect
Why care about green data centers? Data is growing really rapidly for both structured and unstructured data. 2007 Computerworld --- “corporate data growing fifty fold in three years.“ If you’re a consumer, your data lives on your hard drive. If you’re a corporation, it lives in a data center with thousands of servers running, really cold, really noisy. Energy costs in running the servers and keeping it cold. ‘If you spend $1 on a data center $.50 goes to powering the cooling.’ How do we deal with the energy crisis? Can’t really handle the ”more energy“ side of things, so as data center people focus on how to make it use less energy. Part is making sure the software takes advantage of the efficiencies in the server.
What can Data Center Managers Do? Data Center Redesign - energy efficient building (airflow management, electrical structures Hardware Upgrade - energy efficient servers - energy efficient storage Software - leverage the hardware features - be more efficient
Must combine all across industries to address the energy crisis.
Symantec - looking at managing the server more efficiency - most of the time servers are utilized 10-15% Can we use the servers we have more efficiently, so we need less servers to run more applications. - these servers connect to huge storage arrays, which need to be turned on anytime they’re used even if they’re not being fully used - identify which are being partially used and optimize that - make the software recognize the new technologies, like the power management on Intel’s chipset But if the software is always waking it up, it’s not taking advantage of that
Server Efficiency through Virtualization -consolidate multiple servers into one physical machine, but virtualize so it can be treated as separate servers - Symantec has technology for moving from physical to virtual and monitor the workload to dynamically move them around with view to using less physical servers
Discovering Unused Storage
Storage Tiering - sure, we have efficient storage mechanisms, but some have critical information that has been stored in old arrays and cannot afford to switch over to the more efficient models. - Storage tiering lets you classify your storage tiers as efficient or energy-sucking, and this software allows the gradual movement of critical data from inefficient to efficient tiers
Efficiency De-duplication & Single Instance Storage---for example, the same presentation within a large corporation might be within everyone’s machine. What if we store only one copy rather than thousands, and just let you retrieve it
Altiris - partnerships with Intel and Dell to leverage the hardware solutions in the software
Allyson - Intel’s green data centre - start with the microsprocessor - ”processors take up on aggregate about 30% of the total power consumed by a server“ - The Tick-Tock model ”drives energy efficiency“ - each advancement either shrinks existing technology - or introduces new advancements (eg, being able to turn of cores not being used on the fly) - we have performance following Moor’s Law, but the power reduction has been going down ”over a million times“ (per transister) - the power reduction has been very important to support the increase in computing power
How to measure efficiency? - Theoretical energy efficiency looks at either idle or max performance. But servers generally are *not* operating at either of tehse levels. - want the ”real world efficiency“ - want a linear relationship between the power consumption and the utilization, so efficiency is how close it gets to that - did a comparison between 2004 and 2008 servers providing a commensurate capacity (in business operations), over 80% reduction in floor space, energy costs
Jane - data centers focus on ”managing power in an integrated fashion. . . . Data centers are becoming the factories of the 21st century.“ - have seen software as service shift, now with cloud computing seeing data resources shift toward a service model - built virutal world for data centers in different locations in order to monitor hot spots and flow of data - investing in new chip technologies (focused on carbon nanotubes, quantum computing, and something else which are expected to be mainstream around 2020) - reaching the limit in CMOS technology, so need to look at new materials - ”three-stage model to transform the convential data center . . . to a green, efficient, dynamic and responsive data center.“
Simplified: Consolidating fragmented systems to centralized, virtualized systems. Shared: pool resources to get economies of scale. Dynamic: Virtualizing entire environments to get resources on-the-fly
Mobile Measurement Technology - ”measures 3-D temperature distributions within data centers.“ - allows you to monitor and quality-check massive systems in minutes rather than days
Rachel ”Kilo Client Utility Computing Model“ - Kilo Client is a grid computing structure, with 16 nodes - no disks on the nodes, fiber channel connection - try to do fast deployment of may virutalized computers
Brennin - benefits of this system - ”unlimited, rapid client provisioning“ - give you hundreds of copies of images in minutes - ”rapid re-purposing“ - share instances between customers, so ones in tweaking or setup mode that only use one or two instances can be shared with the ones that really need all the servers - ”highly efficient space and resource utilization.“ - if you were doing this, you’d need at least a hard drive per server (and probably more for raid protection). They can do it without nearly as many - ”boot images spread across multiple spindles for performance“ - get your data faster because you can get it from multiple sources - ”simplified back up and bare metal recovery“ - because of the rapid provisioning - ”scalable and flexible“ - if you want more pods, get more pods - ”touch layers of the virtualization stack“ - get backups from everywhere
Aglaia - Industry Efforts - ”all the companies are coming with their own creativity solutions“ - sharing between companies also
Q: What do you do to keep track of where data is? A: Brinnan - have a centralized storage place Q: But once they’re deployed? A: Rachel - software to track who owns which blades A: Aglaia - Symantec has a similar solution
Q: How will the internet data pipes be affected by centralized data centers? A: Yes, will put additional load on the internet. Workload balancing needs to be advanced. A: Aglaia - data compression and encryption is also goign to be critical
Q: What if you have a heterogeneous systems? A: Intel - chose the most energy efficient boxes if you can’t consolidate, ”looking for server instrumentation and management control . . . cooling technologies.“ A: Aglaia - Software that allows for load balancing between disparate servers.
Q: Where in those three areas can we make the most gains? Redesign the racks? A: Aglaia - all solutions have merits. Software gives you quick return, ”but doesn’t really solve the problem fundamentally“. Redesigining hardware ”does help and does not require as much as investment or time as redesigning the data center.“ A: Think about runnig applications not necessarily at top speed. ”There’s a whole new area of researching looking at the trade-offs between optimizing power consumption and [how fast] the applications run.“ A: Cooling is required per processing, so reducing processing also reduces cooling requirements. ”It really is an ecosystem environment . . . you make a change somewhere that you think is going to be good and it has effects other places. . . . We need innovation along all vectors.“
Q: Standardization work? Abstractions like the cloud? A: There are consortiums like ”The Green Grid“ looking for the industry as a whole. A2: ”The Climate Savers“ computing initiative looking at ”inside the box“.
Q: [missed it] Real applications? A: Looking at the Smart Grid, providing information like how raising the temperature in your home affects the energy savings. ”How do you change behavior? Not just for consumers but also in the data center?“ Intelligent buildings and smart buildings are going to be important.
Q: Reliability factor when turning machines on and off. A: Intel - the useful life of the server is less than the microprocessor’s, so not an issue on the microprocessor side. Also need to look at the carbon costs of the machine itself, so yes reliability matters. Also looking into recycling programs and the chemical costs.
Q: Solar energy? A: Kilo using outside air to cool, so similar idea. A: Intel ”largest purchaser of renewable energy in the US“ A: IBM investing in solar power research and new rocesses to manufacture solar sales. A: Kilo uses reusable water.
