Datacenter

You are currently browsing the archive for the Datacenter category.

Rightsizing Servers to Achieve Cost and Power Savings in the Datacenter

December 16, 2009 in Architecture, Datacenter by Julien | 1 comment

Rightsizing Servers to Achieve Cost and Power Savings in the Datacenter: This article introduces a whitepaper (pdf) published December 09:

In a nutshell, the paper describes how we perform detailed analysis of our internal workloads and then select balanced servers that are optimally sized for our production scenarios. It is my hope that IT teams in other companies can use the information in our paper to justify devoting resources to characterizing internal workloads, because that is the basis of an effective rightsizing strategy.

The whitepaper makes a parallel between the servers and the cars purchasing process is interesting as it enlarges the vision on the all criteria to look for, but it just dismisses that if you have an outsourced services or if you just lease your servers, then your needs are not the same and you tend to remove some criteria from your process.
The document goes through SPEC (Standard Performance Evaluation Corporation) and TPC (Transaction Processing Performance Council) benchmark with their respective applications and limitations.
After an overview of cpu, memory, disk and network performance evolution over the last years and the links between theses components, the power factor is introduces through the PUE perspective, highlighted by this figure (from source document):

Three-year total cost of ownership of a basic 1U server

Three-year total cost of ownership of a basic 1U server

The whitepaper continues by studying cases for typical servers roles: web server, file server.

What I find really surprising, is that it never deals with virtualization.

Update : One more very good article on this whitepaper from GreenM3 : ex-Intel engineers at Microsoft share processor secrets, optimize performance per watt

Iliad Entreprises opens new data center in France

December 13, 2009 in Datacenter by Julien | No comments

Iliad Entreprises datacenter under load test

Datacenter during full capacity load test

Iliad  bought a 22 years old data center in Paris and refurbished it into a modern Tier III (not sure it is a certified one, see Tier Standard: Topology) by changing nearly everything inside. Unveiled December the 3rd, final tests and technical presentations were filmed and showed in some french IT news. With a capacity of 700 racks split in ten rooms, it integrates from scratch cold aisles (20°C in the cold aisle, 30°C ambiant) and free cooling with a smart load dependant energy management system (cooling & electricity).  Cooling is based on traditional ice water and use raised floors.

The whole series of articles that popped up are interesting because, it is a recycling operation, tests (load, backup, etc.) are covered and explained by hosting manager Arnaud Bermingham.  All designs have been based on traditional solutions, optimized to provide the best efficiency possible, all year-long. This is why free cooling is present, but efficiency goals are not counting on it, as it is not always available. Therefore, the data center is efficient all time (as said).

I noted the specific points:

  • Electricity and cooling management systems were developed in-house. They created two redundant controller, each developed by a different software engineer to avoid software bugs.
  • Human has no action on the energies management (cooling, electricity), only monitoring access (bypasses, backups,… are all automatic).
  • To reduce rollouts costs, they used temperature sensors on a bus (chained on cat5/6 like cables). It allows the deployment of more probes in order to gain more granularity in measurement. Power consumption is measured at the circuit breaker of each rack departure (two redundant per rack), this last system being as well in-house developed).
  • In case of UPS failure, the data center can be directly fed from the grid, bypassing all securities, if they fail.
  • French links:

I have two comments:

  • I wonder if internal design and engineering are sustainable on the lifetime of the data center, and especially for the maintenance (risk, cost) and evolution aspects.
  • The cold aisle temp could be higher (up to 27°C), if following the ASHRAE recommendations.

This article data and pictures are from ITespresso Datacenter blog and PCINpact coverage.

Week 47 Links

November 20, 2009 in Datacenter, GreenIT, Links, Virtualization by Julien | No comments

Notes from week 47 rss reads.
This week : Cisco, Cloud Computing, Datacenter, Virtualisation and Web.

Read the rest of this entry »

The Data Bunker Boomlet: Data Center Knowledge has a list of bunker data centers. Royal Pingdom features two of the listed in this article: 8 ways to make data centers less boring. As The Bunker is announcing a second facility in London area, this is a trend to follow.  Not mentioned in the list, Telehouse Europe will open its third site Telehouse 3 : Magny (pdf) soon in Paris area (map) in a former secure military place.

July 15, 2009 by Julien | No comments

Data Center Designs (continue)

July 14, 2009 in Datacenter, Virtualization by Julien | No comments

To complete the brief news I posted in April: Top of Rack vs End of Row Data Centre Designs, Cisco has a document from 200902 : Data Center Top-of-Rack Architecture Design (pdf).  Related to the Nexus 5000 switches, it highlights the benefits of “unified networking”, from the physical infrastructure, to the logical one. For those interested in the physical benefits, the document Power and Cooling Savings with Unified Fabric (pdf) extend the discussion to power and cooling. You never get enough green factor.

Edit: ViewYonder has an article “Cisco UCS dog food tastes nice” on cabling at cisco with a slide that says it all.

NYT : Data Center Overload: Great New York Times article coming with a slideshow on datacenter evolution, from the wtf? to usage, cost and strategy/impact for industry (Karl Marx involved).  For any architect, the “Latency concerns are not limited to Wall Street; it is estimated that a 100-millisecond delay reduces Amazon’s sales by 1 percent.” is priceless.

June 12, 2009 by Julien | No comments

Jülich Supercomputing Center (JSC)’s Clustercomputer JUROPA: Sun’s Marc Hamilton has a series of posts (1, 2, 3)on the  where Sun is providing two systems : Constellation high performance computing and Lustre storage system. Architecture is presented and pictured into details up to cabling and cooling.
NASA’s Nebula (Sun Lustre used for storage) is worth a look as well.
Nevertheless, the story does not tell if it’s green or not.

May 29, 2009 by Julien | No comments

Cisco’s blog post A More Granular Approach To DC Cooling, is mentioning two tools:

Data Center Assurance Program: Behind the wording, are standing very useful and strong documents like the DCAP System Assurance Guide 4.0 (PDF – 21 MB) and some Datacenter designs guides:

Depending of your needs, these documents can be too technical or too application oriented (solution for one vendor). I haven’t find Unified Computing System related content there.

Second, the Cisco Power Calculator (requires CCO login) enables sizing and checking PoE configurations.

May 28, 2009 by Julien | No comments