February 21st, 2012 | 
Author: Mon complice en affaires, René Breyel, a quelques mots à dire sur la gestion des centres de données (DCIM). Par le biais de l’AQIII il participera à une table ronde pour en discuter et mettre à disposition son expertise. Plus de détails ici. http://tinyurl.com/75ruvyn
René est certifié « DataCenter Operational Sustainability » par l’Uptime Institute (ATS) et il est présentement en mandat pour le développement et mise en place du Cadre de Gouvernance relatif à la gestion et aux opérations des Centres de Traitement Informatique de Desjardins.
Red lights are flashin’
There’s been a misunderstandin’
- From the AC/DC song : Danger
Like many others, I had bought into the idea that DC powered servers and devices in the data center are much more efficient, power-wise, that their counterparts running on AC.
Well… it looks like we were misinformed by studies based on false or erroneous assumptions. This eyes-opening article by Kevin Brown debunks that idea by showing that by using the best design for both approaches, the DC advantage is negligible.
Another myth busted. No magic wand there.
Hewlett Packard new “Project Moonshot” is about extreme low-energy server technology. A video is here.
Now, that could be a game changer; possibly at the same level that virtualization has been in the last few years. Virtualizing many servers on one physical server has helped reduce the energy footprint and with the current pressure to reduce energy usage escalation in the data centers, it had been a godsend.
But we are still building data centers that are power hungry. From what I can see, the approach so far has been to build where you can use some “free cooling” or try to use green (or “green-washed“) renewable energy, but it seems to me that we are not attacking the basic issue of computing efficiency. In a previous post I had simplistically illustrated where power is lost in a server.
As DCIM experts we usually do not ask questions regarding why a certain type of server is in those racks, and with the higher density bringing consumption over 4-5kW/cabinet threshold maybe we should.
So, do we put DCs on a diet or do we opt for bariatric surgery?
To be able to align DCIM (Datacenter Infrastructure management) and Uptime Institute approach on “Datacenter Operational Sustainability” I decided to register to the ATS certification program.
The next 3 days will be the source of shared reflexion about Tier classification and best practices to support the Datacenter operations.
Stay tuned for the next post !
René B.
|
An interesting report was published in September 2011 by DatacenterDynamics. It combines census data on investments in 2011-2012 with GDP data from the World Bank. Covering a whole bunch of countries it shows the expected growth in data center facilities. It was then followed by another shortly after that showed the amount of energy needed to support such I combined the information from both reports to make my own chart. To explain such surprises as seeing Turkey and South American countries right at the top, it should be understood that the percent in growth is against the Gross Domestic Product of the referred country; so in terms of monies spent, Turkey’s 60% does not even come close to the 9% in the USA. By looking at the expected energy needs, you can appreciate the size of the forecasted investments. If I understand the data correctly though, this gives me an idea of how much emphasis each country is putting on expanding or modernizing their IT infrastructures. In other words, there might be a real push in some countries to get with the times. Either by building new facilities or upgrading older ones.
Looking at the energy demands, however, I remembered another famous report from the EPA where it was predicted that the energy consumption by data centers would double from 2006 to 2011. It had indeed done so from 2000 to 2005. The EPA looked at the, then current, energy efficiency trends and forecasted the doubling of the power usage if nothing would change. But, something did change. From 2005 to 2010 the energy consumption has increased by “only” 56%. A study from Jonathan Koomey shows that the upward trend has actually slowed down. What is the cause of the good news? Virtualization and improved efficiency efforts have certainly paid off and we should definitely continue on that road, but unfortunately the 2008-2009 economic crisis has also played a major role in cooling off data centers investments.
And speaking of cooling, you do know (since I have said it before) that the air conditioning of data center represents between 30% to 40% of the energy used to run it? Some optimization techniques, although they might appear a little expensive for the amount of energy they might save up front, should be looked at for the long term since the savings will compound year over year.
|
 |
La notion de DCIM (DataCenter Infrastructure Management) génère beaucoup de discussions en ligne mais reste encore évasive quant à la description exacte de ce terme. Il est donc important de bien comprendre les différents éléments qui entrent dans sa composition.
La confusion actuelle semble venir du fait que « l’infrastructure d’un centre de données » n’est en fait qu’une considération holistique de plusieurs infrastructures existantes mais gérées jusqu’à présent de manière indépendante et parfois très différente.
- L’infrastructure mécanique du bâtiment et de la salle elle-même;
- L’infrastructure électromécanique des systèmes de climatisation et d’énergie;
- L’infrastructure d’hébergement physique des systèmes (cabinets, KVM);
- Différentes infrastructures de systèmes (serveurs, stockage, systèmes spécialisés);
- L’infrastructure de câblage structuré et de transport des données (Lan, Wan, Wireless).
On voit ici que le DCIM se situe entre deux domaines fondamentalement différents soit la gestion du bâtiment d’un côté et la gestion des TI de l’autre.
Du côté de la gestion de bâtiment, les fonctions de type BMS (Building Management Systems) sont connues, exploitées et bien implantées depuis de nombreuses années. On parle ici de la gestion des systèmes de transfert de chaleur (HVAC), des génératrices de secours, des systèmes de détection de feu, de contrôle d’accès, etc.
Du côté de la gestion des TI c’est une toute autre affaire car la poussée constante des technologies et de la virtualisation amène régulièrement de nouvelles contraintes dynamiques (charges de traitement, densité, chaleur) qui imposent une révision constante des modes de gestion.
Pour augmenter la confusion, certains éléments se situent sur l’une ou l’autre des frontières, tels que les solutions de virtualisation (VMWare ou équivalent) par exemple. On peut effectivement argumenter sur le fait que cette nouvelle couche d’abstraction située entre le physique et le logique dépende de l’infrastructure physique du Centre de données ou soit gérée de manière logique par le groupe TI. Un partage de ces responsabilités semble le plus probable.
Sur l’autre frontière, la gestion des unités d’air climatisé impose un autre partage de responsabilités lié à la gestion du bâtiment et à la gestion dynamique de la capacité imposée par les équipements informatiques.
Chacune des infrastructures identifiées précédemment dispose aujourd’hui de son propre groupe de compétences et des processus de gestion qui lui sont propres. Ces derniers sont matures et efficaces pour certains mais souvent mal définis, voire inexistants pour d’autres. Le DCIM a pour objectif de consolider et d’intégrer efficacement l’ensemble du processus de gestion de manière holistique afin de considérer une seule Infrastructure cohérente et dotée de l’agilité nécessaire au support des besoins d’affaire.
Disposer d’une approche DCIM intégrée et efficiente sera donc bientôt indispensable à tout Centre de données qui se respecte.
Pour en savoir plus sur le sujet, n’hésitez pas à contacter l’auteur ou à partager votre point de vue en nous laissant vos commentaires ci dessous.
René
Tags: BMS, Centre de données, DataCenter, dcim, Gestion stratégique, Infrastructure | 
No Comments »
In the last few months we have had a few conversations, my associate and I, regarding today’s perception on managing the Data Center. Even discussions with other experts in the fields have left us wondering why does it feel like we are breaking ground in this area.
So, I thought this would have been a nice topic for a blog ticket, but I stumbled on one from Dan Fry that covers exactly my points and more.
Since I just read it, there is the danger that I will just regurgitate what he says. Please follow the link on his name.
Dan makes the case that DCIM is not just about cooling and power monitoring; it is about Management and basically about everything in the Enterprise.
Possibly the only argument I may add is that there is no such a thing as a single DCIM solution. To us, DCIM is about PROCESS, whose ultimate goal is to increase the data center agility to support business while maximizing energy efficiency, reducing costs and increasing resiliency to outages.
The good news however, is that, after years of neglect we seem to experience an era of DCIM renaissance where the physical ecosystem takes finally place on the center stage, along with the other IT necessities.
Suite à la publication par la firme DataCenterDynamics du rapport 2011 sur l’état du marché international des Centres de données, il est intéressant de constater une augmentation soutenue des projets (7% de croissance prévue pour 2012), induisant une augmentation de la consommation énergétique et un accroissement des besoins d’outils de gestion, de planification et de monitoring.
Quelques chiffres sont intéressants à relever :
La consommation d’énergie des Centres de Données au niveau mondial est évaluée aujourd’hui à 31 GW, soit l’équivallent de 25 millions de maisons (soit tous les foyers d’un pays comme l’Angleterre, la France ou l’Italie). Cette consommation devrait encore augmenter de 19 % en 2012 !
La consommation mondiale moyenne par cabinet est de 4,05 kW (48% < 5kW, 28% de 5 à 10 kW et 14% > 10kW)
Au Canada, cette consommation moyenne est de 4,33 kW / cabinet, avec une projection d’augmentation de 9% en 2012.
Les coûts et la disponibilité des ressources énergétiques restent les enjeux prioritaires des gestionnaires de Centres de Données à travers le monde. Le Canada, la Russie, le Moyen Orient et l’Argentine restent les zones les moins à risque à cet effet.
L’implantation de solutions de monitoring énergétique est justifiée principalement par un besoin de disponibilité des services (56%), suivi de la réduction des coûts (31%) et enfin en raison de leur impact environnemental (13%). Au Canada, seulement 25 % des Centres de Données assurent un monitoring permanent de leur efficacité énergétique.
Seuls l’Angleterre (pour des raisons de taxation) et la Chine (pour des raisons de croissance économique) mettent sérieusement l’emphase sur le monitoring des émissions de carbone (> 31%)
Pour l’ensemble des Centres de Données, le monitoring permanent de la température et de l’humidité sont les éléments les plus surveillés, suivis de la consommation énergétique et la gestion des flux d’air.
La barrière des 5kW par cabinet se confirme. Elle représente le point critique au delà duquel les coûts d’exploitation prennent clairement le dessus sur les coûts d’infrastructure, forcant les exploitants à réviser radicalement la gestion de leurs Centres comme un écosystèmes complexes nécessitant une gestion holistique des différentes composantes.
Pour plus d’informations l’état du Marché mondial des Centres de Données, l’étude de DataCenterDynamics est dispoble à l’adresse suivante : http://www.datacenterdynamics.com/research
For starters, you can turn on the lights!
An article, from Robert McFarlane, makes the case for better lighting along the aisles in the server room.
The ability, or lack of, to properly make out the details of your wiring connections can have an impact on the probability of errors during interventions.
There are good reasons to think about proper lighting right during the design of the room, but unfortunately it is often addressed as an afterthought. Racks, and servers, are frequently in dark colors which does not help.
A nice analogy put forward for the best data centers lighting is to treat the racks as library stacks. In a library you want enough light to be able to read the titles of the books. Here you want the same, with the added requirement that light should be able to penetrate inside the cabinet as much as possible.
And if any concerns about energy usage are raised, especially because most of the time nobody is in the room, that became a moot point long time ago with movement sensors and today’s LED lighting which has reached full maturity.
With their mostly black & white looks, Data Centers irradiate an aura of an aseptic environment. But are they?
Contaminants get introduced in the servers’ room ecosystems many different ways but they maintain a cloak of invisibility for a long time before they get discovered.
And they can cause all kind of problems that, unfortunately, get filed as having “unknown” causes. This is because the link between contamination and hardware failure is often overlooked.
A 2009 ASHRAE paper (downloadable here for free) sheds some light on the issue and provides some “beautiful” pictures of electronic components that have been exposed to these gases and dust.
Dust and other particulate can also cause some other issues not necessarily related to possible electrical discharge but also mess up with fire suppression systems. A rarely caught on video example of this is here: http://www.youtube.com/watch?feature=player_embedded&v=2fyGGqgVzCY (things start to be interesting at the 1 minute mark). The sensors cannot really distinguish between dust and smoke particles. An accidental release of Halon can occur and it has been evaluated that a third of Halon gas dumps are actually accidental. And there is a well known link between this gas and damages to the Earth ozone layer.
Humidity is also another gremlin lurking in the nooks and crannies of the Data Center, often not so much because of any pipes leak but often due to condensation caused by an improperly adjusted CRAC (don’t get me started on the cooling aspect…).
Once rust sets in a sub-floor for example, metal oxide particles could get blown around and possibly land on micro circuits.
And we know that dust sticks to servers fans and clogs servers inlet vents, thus reducing equipment life.
But… these are not the only dirty elements related to a data center; dirty generator fuel is another. If you have a diesel generator, do you keep track of how long the fuel has been sitting in its tanks?
Fuel is an organic substance, thus vulnerable to bacteria. Micro-organisms can proliferate and actually spread through the fuel line, making a sludge non very pretty to look at.
Some pictures and more info here .
So the next time you enter a computer room, I hope you see it with new eyes and don’t get fooled by its pristine appearances. It may not be as immaculate as it looks…
Nouvelles- Lancement de la version 6.2 pour la suite DCIM nlyte
- Intégration du service de stockage en ligne d’Ikoula dans le logiciel Allway Sync
- Global Switch ouvre le plus gros data center de l’hexagone
- CyrusOne change de pilote pour accélérer son développement
- Le Data Center d’Apple en Caroline du Nord peu bénéficiable à l’économie locale ?
