Clever Geek Handbook

Six rules for choosing a data center

Choose a data center for your project - that pain. Finding the best data center for your tasks is easier if you know what to look for when searching. In the case of the Playkey cloud gaming platform, the quality of the streaming directly depends on the data center: its reliability, proximity to the user, network quality and other parameters. Therefore, we work with data centers every day and know their main problems.











In this publication, the technical director of Playkey, Alexey Lykov, will tell how to choose a data center, what not to forget when searching, why delve into the particular power supply and how European data centers differ from Russian ones. We hand over to Alexey.







1. Study the data center Tier standard



The first question when choosing a data center - reliability, or uptime. The most popular certification for this parameter is the Uptime Institute methodology, which divides all data centers into four ranks (Tier). There is an alternative system for assessing the reliability of data centers according to the TIA 942 standard, but it is not so interesting to us due to its advisory nature. In the case of the Uptime Institute certificate, they themselves check the DCs at several stages (from the design of the documentation for the building to the regular evaluation of the operational services of the center) and issue an assessment based on the results of the check. Get yourself such a "fake" certificate for DC becomes almost impossible.







Here is a brief description of each of the four Tier:









A map of all certified Uptime Institute data centers is here , but their location in Russia. A total of 26 pieces, and all of them are located exclusively in the European part of the country:











The blue one on the map is the only Russian data center certified by the maximum TIER IV standard, and it is located not in Moscow, but in Saransk. Here’s how it looks from the inside:











The photo clearly shows the removable raised floor, which is arranged like this:











In accordance with the classification of the accessibility level of the data center by the Uptime Institute TIER IV, it provides full redundancy of all engineering systems with a factor of 2N + 1, which allows for any repair work without stopping the data center services.







Here is the power supply system with independent inputs of electricity for the Saransk data center:









According to official information , the data center in Mordovia should become a central unit for the e-government infrastructure, but it will also cope with the tasks of banks, insurance companies, and even the military.







Requirements of various Tier sometimes seem a bit confused (for example, the need to have separate parking for employees and guests of the DC), but most of them are adequate. So, DC according to Tier 3 standard must have a room in the property. It is clear that if the data center is built in a rented location, then the owner of the building can at any moment come to anything, and he either increases prices, or breaks the contract altogether, whatever the penalties. If you see a site with a Tier 3 certificate or higher, this risk is excluded. For ourselves, we chose to work with a data center not lower than Tier 2, and our European data center in Germany is certified for Tier 3 level.







There was a curious case, when the DC rented a room on the high floor, and the elevator belonged to others and, accordingly, it was turned off for weekends and nights, and it was problematic to start the equipment.







2. Specify the parameters of power supply and cooling data center



If you use a non-standard configuration of servers, then the second item can be safely put on a par with the first. We are always especially attentive to the power supply and cooling of the servers, because we use not ordinary, but GPU-servers, for which most of the data centers are not designed.







If we take a standard server rack of 42 units, then it will normally look like this (laptop is not attached):









Usually, the data center owners supply 7-10 kilowatts of power to such a rack and provide appropriate cooling. This is enough for a design with typical servers that consume about 300-500 watts incl. various storage systems, switching equipment, patch panels and so on. But in our case, only one video card in the GPU server "eats" from 150 to 250 watts, and such video cards use up to 8 pieces on one device. As a result, the total consumption of a GPU server is about 2 kilowatts.







Because of this, we can put only 3-4 servers in one rack, i.e. 6-8 units, if we are talking about 2-unit servers. As a result, in the server rack, we can occupy only 8 units, and 34 units will be free. This raises the question of efficient use of space. And raising our rental prices.







Conclusion? If you have unusual requests for equipment, then you should look for a data center, sharpened (at least partially) for clients like you. In our case, for clients with GPU-servers, with a large supply of power to the racks and enhanced cooling. So far, there are very few such clients (and data centers).







3. Choose location by location of your users.



It sounds trite, but it is worth recalling that the location of your R & D or headquarters does not have to be the place for your data center. The primary is where the bulk of service users are located. In Playkey, the development office is located in Perm, and at the start we also worked with the local Perm data center, which simplified setup and made it possible to resolve any issues with the data center faster. But when the user base began to grow actively, and its main part (of those who play in Russia) still fell on Moscow, we decided to group all the servers also in the center.







After the capital move to the Moscow data center, we immediately saw an improvement in the quality of communication and a positive feedback from the players.







4. Look for management contacts and “test” technical support in advance.



Rapid technical support and adequate guidance is another crucial point in the work of the data center. Try to clarify a typical task with the data center employees, for example, set up according to the BIOS instructions, and look at the reaction. If you are sent to another specialist or start to slow down with an answer, all these are not very good signs. Remember that if something serious happens to the server, the issue is most often not solved without physical presence. And tech support - your "hands" in case of any serious problems.







As for the management of the data center, remember that without the contact of the "decision maker", any coordination will take long days and weeks. If you do not meet the person who actually makes decisions in the data center when signing a contract, it is better to look for an alternative platform for hosting servers.









5. Check the legal entity of the owner of the data center



The shortest, but no less important point. Do not be lazy to "punch" a legal entity on nalog.ru. Look at the extract from the Incorporation - and you will understand how long the company has existed, how often the management and owners have changed. And in the base of arbitration cases do not forget to see her court cases.







6. If you work with a European DC - be patient



The mentality of European data centers and their organization of work are very different from Russia. In our experience with European data centers, we can say for sure that they are doing more and more slowly. If in Russia, technical support employees can quickly solve almost any problem with server settings, in Europe everything is not so simple. They always have an internal division of employees into those involved in low-level and high-level tasks. The bureaucratic machine forces customers to prove that the problem is really important and requires the immediate intervention of a "high-level" specialist from the data center. Get ready to describe in detail the reason for your appeal, wait for approval "with the tops" and only then hope for help.







There are differences in the approach to work. For example, in Russia, almost all requests to the data center can be sent via chat or letter and promptly receive a response. When we work with employees of data centers in Germany and the UK, we are forced to communicate only by phone. Why? It's just that they have made it this way.







Such nuances make Europeans even slower. It is worth making a discount on the fact that we ourselves are from Russia and it is easier for us to interact with the Russian-speaking support, but even without this, the processes of “ours” are, in my opinion, lined up better and the work goes faster.







We have now begun preparations for entering the US market, so for us all the points described are far from an empty sound. With the states, we will be located no longer in three hours of flight from the data center, plus there will be a decent difference in time zones. So we will be very responsible when choosing a long-term partner. A plus for a quick start will be a cloud provider, since over the past year two major services have already launched the ability to rent servers with a GPU.







And for those who want to check the quality of our data centers, we have prepared the next PlaykeyGeekSummer promo code for 720 minutes of the game. It works for four days from the moment of activation, suitable only for new users, and do not forget to activate it until July 31, 2017.







More articles:


All Articles