Obsolescence of data centers and servers, a matter of availability

The interest of a company to put its servers in data centers based primarily on economic reasons : eliminating the costs of air conditioners, UPS, security access, maintenance of the room etc …. However it is necessary to remain vigilant on the quality level of service to which it entrusts its servers. Is it useful to remember that for most companies the « blocking » of its information system means a complete cessation of its activity ?
A recent study showed that 60% Failures of computer servers are due to failures from the environment wherein they are installed :
– Breaking of the electrical network, automatic or without sufficient recovery by UPS,

– Failure of air conditioners, transforming the computer room in a « sauna »,
– Dust deposition, generating false contacts in the circuits of servers

Originally many datacenter were built from the refurbishment of office buildings under the pretext that they already had a computer room. This leads to a rehabilitation facility originally healthy, but quickly reached their limits of reliability as and when they are « filling ».
Beyond an occupancy rate 70 %, it is common to see appear amenities that erode the reliability of a data center, for example :
– Creation of private rooms but only one air conditioner or without backup UPS lack of space
– Reducing the number of test operations and maintenance due to access difficulties : air conditioners in the rooms berries, rooftop
– Poor distribution of cooling air flow after changes in the arrangement of rooms : deletion or addition of partitions
– New equipment rooms without duplication of electric meters due to wiring problems or siteThese disparate changes have not only led to a decline in the quality of data centers, but also more complex daily operations : multiple interventions, difficulty of access, absence of substitution solution during interventions. Besides, it is common for operators to give up announcing the closure of their old datacenters.So what should be done ? Resume its servers. Not at all. You just keep in mind that outsourcing server is a project that falls within the period. It is entrusted by its servers for 12 month. One should choose a provider capable of providing consistent operating environment for 10 years to come whatever its filling rate, changing technology computer servers, operating costs … a datacenter XXI century.

The Marilyn Datacenter keeps its promises

One year after the commissioning of the datacenter ‘Marilyn’ of the ISP to Internet CELESTE, It is an opportunity to revisit the first results of its operations. We interviewed Nicolas Aubé, President of Marilyn

M. Aubé, can you remind us why this datacenter is innovative ?

NA : This is the first ecological computer center and high density. Its vertical design allows a cooling by ambient air almost year-round. The hot air generated by the server is used for heating our offices, the surplus is discharged via automatic shutters at the top of the towers.

Why this name of Marilyn ?

NA : Because we thought the famous actress in the movie ‘Seven year itch’ when her dress flew on an air grille. In our datacenter, the air passes through the floors and can soar dresses !

marilyn-nicolas aub

Your project aroused the interest of the public ?

NA : Yes, We were the first surprised. We did a first session open door which has been so successful that we have fact… 20 editions. We have had many reports of press paper/Internet and national television channels. I believe that the hidden face of the Internet interested public.

Is the operation optimal year-round ?

NA : Since commissioning in October 2011, in marne-la – Vallée, We had the opportunity to test extreme outdoor temperatures between-15 ° C in February and 40 ° C in August. Ambient humidity ranged from 15% to 100%. We were able to validate all of the beaches of functioning of the datacenter.

What are the modes of operation ?

NA : Cold mode is when the temperature is less than 22 ° C, either 80% of the year. There not need air conditioning to cool the air. We recycle hot air discharged by computers and mix it with cold air, outside, in order to blow air to a temperature between 17 ° C and 22 ° C in the cold columns before the servers. The cooling mode is so effective that we had to limit it the first weeks, front rooms are filled, because the temperature tended to be too low.
A very good surprise is also functioning in the ranges between 20 -22 ° C : cooling by ambient worked well even in these relatively warm temperatures : the temperature in the rooms was not higher than the outdoor temperature, Despite hundreds of computer servers in operation. And this without any air conditioning. This is the heart of our model : We only spend the energy to turn computers, not for air conditioning.«Warm» mode is between 22 ° C and 32 ° c.. Here, We always use the outside air, because its temperature remains lower than expelled by servers. We climatisons the outside air to bring it back to 22 ° C, and we repel any warm air by the roof. This mode remains effective enough.
The ‘hot’ mode occurs above 32 ° c.. We then operate closed circuit, cooling the hot air produced by the servers, as a regular data center. In this mode, We don’t realize economy, but it represents only 5% of the year. We have noticed also that it remains interesting to partially open the flaps at the top of the towers, to create a depression in the hot aisle and a natural drawing effect.

What are the constraints of the process ?

NA : It requires a tight seal between the front of the servers (cold aisles) and the rear (hot aisles). For this, We provide our clients with caches. We found that in the case of too much fresh air leaks, We increase the pressure in the cold aisles.

How have your customers they accepted these innovations ?

NA : We were a little worried on knowing if our customers accept this model of vertical computer room ; and even if some would not have Vertigo because it works above 5 floor gratings. However we are pleased to see that this is not the case, customers immediately joined our concept and these innovations have been rather a commercial engine as a brake.

Why do you say that the Marilyn datacenter is in high density ?

NA : It’s available for Bay power. Through our process of cooling, We provide berries in high density without any modification at our facilities. The datacenter is equipped with a medium density 5 kVA per Bay, What is significantly higher than the other datacenters we know in France. We provide standard arrays up to 10 kVA.

Customers need this high density ?

NA : It depends on the type of equipment and use. We find that customers who use servers blades, making virtualization or cloud computing need high-density. However they need less space. One of our customers had a suite of suite 10 berries with 20 kVA in a datacenter in island of France. He chose to move in CELESTE taking that 2 high density arrays, with 20 power kVA. He was able to achieve savings in the management of its facilities and its data center budget.

As with other data centers including one in Courbevoie, have you suffered outages this year ?

NA : Non, the datacenter has been available to 100% as we guaranteed it to our customers. Indeed, We have two completely independent electric strings, in separate buildings, Since the high voltage to the distribution in the bays.

Can you tell a client reference of the Marilyn datacenter ?

NA : For example Econocom, a leading European integrator. They have been seduced by our innovations in terms of reduction in energy expenditure. They have trusted us since the opening of the datacenter.

Have what changes you made since a year ?

NA : We have extended the commercial range in terms of density. Indeed, We were able to respond to calls for tender seeking up to 15 kVA per Bay. In addition we also put in the catalogue of the berries with 1 kVA for low consumption requirements. We launched berries « cloud » with variable consumption, charged to the use. This offer is ideal for Cloud computing providers. We have also exploited the synergies with our range of fiber 1 Giga. We propose an offer coupled with a Bay in our datacenter Marilyn and a fiber optic to 1 Giga Ethernet VPN client.

Concretely what benefit this offer brings to the Manager of an undertaking located in Paris for example ?

NA : He moves his computer in a green datacenter. It has access to its remote servers as if they were on its local network of Paris. He won the m2’s Parisian premises, decreases its electrical costs and enhances the security of its systems. If it is the choice of Marilyn as a back-up site it has a high availability and redundant infrastructure.

What are your plans for the future ?

NA : En 2013, We planned to equip the second tower of the Marilyn datacenter. Furthermore, CELESTE is a member of cloud, the collaborative research project on cloud computing, which is supported by the State ecological. In this respect, We are working on a model of data centers network, dedicated for the cloud. We have also been approached by investors or manufacturers who wish to reuse our patent ‘Marilyn’ in France or abroad. We have designed a license therefor agreement.

In which countries the datacenter Marilyn could be replicated ?

NA : in many countries ; If the climate is not too hot and dry. Everywhere in Europe, except South ; in the North, the centre or East of the USA, in the Brazil, in Southeast Asia, at the Japan, in South Australia.
Nicholas Aubé, Thank you

Marilyn datacenter, a 100% availability warranty

Just after 10 months of work, construction datacenter Marilyn CELESTE ends for commissioning in late September. Technological choices made throughout the project are now Marilyn the most innovative datacenter : ecological, high density with availability 100% and the power supply for maintaining the temperature.

Last October, the first stone of the computer center is placed CELESTE ; and 10 months after the datacenter is already almost finished. This new concept of datacenter arouses curiosity, panel visible presentation of the A4 has much. Visits are increasing daily : Hosts, banks, SSII or curious of all data centers Ile de France…

I must say that the technical choices make CELESTE Marilyn the most innovative datacenter year

  • Double electric chain with arrival ERDF 3MW
  • Two HV 20 000 In
  • 2 BT with local
    • tables with service levels “3 3 3″ to perform Maintenance, room change or upgrade hot
    • two inverters with inertial component thus avoid cool battery
  • Electrical distribution doubled for each bay
  • A generator sized to resume full power center

In conclusion, this architecture design TIER IV ensures availability of 100% power supply and temperature maintenance with a simple and robust approach. This is what attracted visitors beyond the quality of the suppliers selected for berries, inverters or energy efficiency very low 1,3 (PUE).

Originally patent Marilyn and President of CELESTE, Nicolas Aubé is surprised and delighted by the enthusiasm of the community to Marilyn “Requests for visits and solicitations conferences worldwide are increasing and the rate of pre-booking is higher than expected. We are proud and excited at the same time be 26 September for the first breaths of Marilyn.”

Fundamentals of High Availability

High Availability concerns more and more companies as individuals, by dependence that create the Internet and new technologies, which are not always available. There is no standard regarding the duration of service interruption. It depends on the context and the criticality of the application.

For example, a navigation system in an aircraft is designed to have a lockup period 5 minutes per year, while the application billing company will be designed to a lockup period of one day per year.


haute-dispoHigh availability is defined as a system to ensure operational continuity of service over a given period. To measure the availability, a scale is used which is composed of 9 levels. A Highly Available Service is 99% available less than 3,65 days per year.

Calculer affinity to Availability, the following metrics are used:

1.MTBF (Mean Time Between Failure) : measure the estimated time between 2 failure of a system.

2.MTTR (Mean Time to Resolution) : measure the estimated time to restore functionality.

At East formulas availability : Availibility MTBF = / (MTBF + MTTR)

Internet and High Availability

As more and more businesses, Internet is at the heart of the business and the need for availability is constant. Indeed, This media is used to communicate as much outward support for many business applications (CRM, ERP, etc.) or telephony.

It is therefore necessary to distinguish between the needs of the company on two levels : services available to customers versus services necessary for the internal functioning. One of the most telling examples is the websites of companies, which are now at the center of communication and most business enterprises.

High availability websites is organized around different axes which can be crucial:

• redundant hardware,

• Location of equipment,

• Application updates server security applications,

• securing the corporate network,

• the continued availability of backup solution / emergency / disaster recovery,

• sizing power equipment.

Hardware redundancy

Redundancy is the mechanism to replicate one or more components of an architecture with one or more identical elements.

Have n server(s) sur x site(s) allow redundancy of information, with a risk of failure divided by n x…

However we need systems that can automatically switch from one site to another. Systems most commonly implemented to ensure this redundancy are clusters.

Clusters can be active / passive or active / active. The first case represents a group of machines on which relief toggle infrastructure, while an active / active system will have both systems in parallel operation; in that event, one of the two materials can work solo.

Maintenance of applications and software updates

Applications can submit bugs, the resolution updates can correct these defects. Thus we can avoid malicious people explore a loophole that would allow access to company information. Have maintenance service therefore is important and sometimes taking into account the technical skills, it makes sense to outsource maintenance.

Failover in the heart of the device

It is a plan that allows you to resume full or partial activity after a disaster occurs in the information system. The purpose of this plan is to minimize the impact of the disaster on the activities of the company.

The key points in a recovery plan are:

• backup equipment

• the availability of emergency equipment

• backup solutions, with degraded mode (service quality) ex.: backup link with a smaller flow

Enforcement : the availability of internet for business

More and more applications require high availability of the Internet to function… The functioning of the Internet (see article on lab CELESTE) but also specific Internet connections help ensure full availability of it « naturally ».

Advanced solutions can greatly reduce the risk of failure or degradation of services :

1. Have multiple internet connections through different land issues

2. Ensure the permanent availability of a backup solution / emergency / disaster recovery (mode transparent)

3. Opting for connections with Recovery Time Warranty

4. Choose connections with guaranteed bandwidth

Have no internet connections

Have two internet connections via two different land locations may allow secure Internet access solution. In case of failure of one of the lines, This traffic is routed to say is automatically redirected to the second. The presence of two routers in active / passive mode strengthens the redundancy of the system. In that event, it is better to opt for an automatic backup, transparent to users.

Have a backup plan

In case of hardware failure, the material can be redundant infrastructure operator : while the equipment works seamlessly in case of failure, and as before the implementation of two routers system redundancy reinforces.

Opt for a warranty Restoration Time

The GTR is the guarantee of recovery time on a connection in the case of a service interruption. The GTR must ensure that the service interruption is the least detrimental to the company.

A GTR 4 hour internet connections will be the ideal option for IP telephony solutions or as part of a corporate IP VPN , especially when it allows the use of an ERP / centralized CRM.

Choose connections with guaranteed bandwidth

Even if the Internet service is not completely interrupted, it could be greatly altered. In that event, should check with their service provider to have a guaranteed rate. This is particularly important in the case of IP telephony. Indeed, an alteration of the quality of the link will direct consequence a decrease in the quality of the telephone communication.

For all links, SDSL, optical fiber but also ADSL, to have a guaranteed rate, a priority channel to ensure a minimum bandwidth for each specific application or use of the company is configured (telephony, internet…).

To provide a highly available service, must ensure that the infrastructure for the provision of this service is functional 100% time. In this article we are mainly focused on internet links, but do not forget the energy, air conditioning, servers, etc..

Criticality or availability rate necessary for an application or service thus guide the choice of Internet connections to implement a high availability solution. And all connections are not worth !

The R&D service