How to choose a server for business in Europe

The server is a key component of the IT infrastructure and corporate network. Its configuration is a complex task, you need to know vendors, model lines, platforms, components in detail. For example, only HP platform offers many options for different tasks, not taking into account the possibility of combining case and motherboards.

How to choose a server

Before you buy a physical server, you need to determine the main and additional criteria from which to start choosing.

To simplify the process, we created a classification with no regard to hardware capabilities and application scope. We focused on finding a balanced option for solving the tasks set for the most common hardware components that can be found in all vendor configurators.

The choice of equipment is closely tied to reducing the company’s costs, customer trust, data security and how the equipment will cope with loads. Some companies need a simple office server is enough, others want a high-density superblade server with other servers inserted in it.

How to choose your perfect server?

Know your goals

The main criterion should be your goals: why are you buying a server, what tasks and roles are fundamental:

• high performance (for example, terminal servers are used to provide remote employees with access company resources, SQL/DB are for information processing and subsequent storage);
• large volumes for content storing (file servers are for organising and structuring information, web servers are for hosting sites, services and services);
• cloud computing.

Simple server models may lack an IPMI/IDRAC/ILO port for remote server management.

Additional criteria help to determine the optimal combination of components, their functionality and cost of both purchase and further operation.

Chassis type

When choosing the type of chassis, you need to take into account the number of components placed inside the server (placement options in different form factors, ways to combine the server, peripherals and equipment in one case). The larger the size, the more disk space you need. For example, a server to grow into starts from 2U and more.

There are racks, towers and blade servers. Understanding the chassis type is necessary to optimize the physical space. More detailed classifications include:

• classic servers with 1-4U;
• blade servers (4-8U) with the main disk subsystem located outside the main case since it is weak by default;
• high-density servers with many processors and processor cores within one case or a large number of cores in a server rack (42U). Servers have a powerful disk subsystem and are a cheaper alternative to blades;
• servers for graphics processing with many graphics cards, a powerful cooling system and a small disk system (if compared to the size of the case);
• data storage systems based on a server platform with 1-4 processors at the physical level, RAM and the ability to install a huge number of high-capacity disks. This solution is used in data centers, service providers and clusters of mail servers for cheap storage of data that does not require constant access.

Disk subsystem

Disk subsystems are differentiated:

• by the number of slots and purpose;
• built-in (SAS HDD/ SSD);
• NVMe;
• hybrid (SSHD): combining SAS HDD (for data storage) and SSD (as cache memory);
• external/with different types of controllers on board/with expansion options.

On most servers, hard drives and SSDs are combined.

Hybrid operation is a function of the disk subsystem controller. Fast disks are used for frequently accessed data (according to the controller’s analytics). The data is then replicated and transferred to slower drives. When a change to the SSD occurs, this data is slowly cached to the SAS HDD.

How to determine a subsystem you need? A simple 1U server with ten disks is suitable for a small office with up to 100 people. But its performance will immediately decline as soon as the number of users is more than a hundred. In this case, the company will need at least a two-unit server. If the company wants to purchase a multifunctional server and has more than 100 employees, it is better to buy several servers or a slow backup server to protect the business from downtime, reputational and financial losses in case of equipment breakdown.

If you need an all-in-one solution, then you can choose a compromise option for all the components of the server.

Performance, reliability and disk space depend on the types of workloads.

• To manage the periphery, a low-power server is sufficient with a small amount of RAM and a minimum number of processor cores.
• For a website, the selection of hardware and software components depends on throughput, traffic and resource structure.
• An application server requires more complex functionality, many cores and a large amount of memory for fast calculations.
• A mail server runs on dedicated servers with a reliable disk subsystem to store large amounts of data.
• Game and search servers need a lot of computing power.
• A server for SQL and databases provided to other servers needs to have a high-performance disk subsystem, a fast multi-core processor and significant amounts of RAM.

The disk subsystem sets priorities: what is more important for the server – the speed of searching for information, the time of processing requests, the volume of media, the cost. Hard drives (SATA3, SAS, SSD, NVMe) differ in spindle speed (or speed of reading data from flash memory), volume and connection interface.

Whichever interface of the disk subsystem you choose, give preference to those that have the highest total throughput. All elements of the disk subsystem (disks, disk controllers, disk connections on the board) must have maximum performance.

There are general recommendations that you should pay attention to.

• Availability of integrated devices (SAS HBA/RAID, 10 Gbps network controllers (if there is no need for redundancy, it is not worth paying for it).
• After the server is assembled, the slots in empty processor sockets become unavailable, as well as integrated devices. All elements are counted and installed before the server starts.
• Servers optimised for GPU cards for high-capacity computing should include additional power supplies and heat sinks.
• To meet the standards of high fault tolerance and redundancy, it is necessary to use double power supplies.
• Servers for telecommunication racks or cabinets with low depth do not have a rail and are installed with lugs on a 19-inch plane or on a shelf (optional for this type of racks).

Type of power supply unit

Power supplies work 24/7 and withstand hard loads. The difference lies in the combination of power, type and length of cables, cooling system, number and types of connectors, design features of power supplies. In total, there are six categories based on the prestige, quality of components and assembly (but not the performance). The following colour specifications are distinguished:

• white (basic);
• bronze;
• sliver;
• gold;
• platinum;
• titanium.

If you need high fault tolerance, purchase power supplies not lower than the golden category. Also, the fault tolerance depends on the number of power supply units (1 or 2) provided in the server. It is not feasible to buy a power supply with an excessive power reserve (more than 30-90%), it is more profitable to invest in better components and efficient cooling.

The number of users is a critical factor in determining the type of processor and the required amount of memory. It is necessary to understand:

• how many user sessions can simultaneously be on the server and how many requests  are generated on average over a certain time;
• what is the peak traffic;
• prospects for increasing the number of visitors to calculate the volume of network servers.

For some software and hardware platforms, there are basic recommendations and standard formulas for calculating created by developers based on the allocated amount of resources and power for each user.

For ERP systems with terminal sessions, the following formula will be valid. Each user is allocated a minimum of 2 GB of RAM and 10 GB of disk space, as well as one processor physical core designed for a maximum of 10 users. If you need a server for 250 users with a peak load per unit of time, then you will need 25 processor cores, 500 GB of RAM, 2.5 TB of data. In this case, everything will be running and backuping smoothly and users will not suffer from a lack of space for data storage.

The choice of server has to take into account your budget limits. The more complex the tasks are, the more expensive the server is.

Cheap server:

• does not work autonomously;
• is not designed for critical loads;
• has no redundancy;
• doesn’t work with a lot of traffic.

Expensive server or cluster of servers:

• works 24/7;
• is designed for complex calculations and critical loads;
• performs redundancy functions;
• handles critical traffic, is responsible for mission-critical processes, services.

There is no perfect server that is suitable for a huge range of tasks.

Server performance is directly affected by the type and performance of the processor system (the number of physical processors, the number of cores in them, the frequency of the cores, the amount of cache). The main market offers for medium and light servers are presented by Intel Xeon and AMD. You can choose a specific architecture only after analysing the tasks that the server will be used for.

RAM for servers is expensive and has a significant backlash in performance. The amount of memory depends on the type of system and its requirements.

To select a sufficient amount of RAM, you need to consider the type of bus on the processor and the number of channels (2 or 4). The number of channels depends on the type of processor.

The compatibility of the processor socket depends on the chipset on the motherboard used to connect all attachments. The memory clock speed depends on the processor. The presence of additional controllers on board determines the stacking of external memory modules of the disk subsystem to the server.

For example, if we need a slow server for backup, we can buy a cheap one-unit solution and an inexpensive shelf to it with a power supply and disks connected to the server through a controller. As a result, two inexpensive servers, one of which is actually an empty box, give a lot of relatively fast disk space for backups for little money.

Server and provider selection

Many data centers stick to standardisation and the choice of servers is limited to typical configurations. Customers can make calculations for their optimal platform using the configurator, but few data centers will provide custom servers for individual projects. You have to independently search for contractors, purchase equipment and arrange delivery with a logistics company.

The best provider may be the one who is ready to fulfill the wishes of the customer, take on all the work to provide a non-standard server. The one who will help to calculate resources, configure server equipment taking into account all aspects of the project. For different needs, you may need both a server and separate disk shelves, switching systems or network protection. In addition, the provider will order, deliver and perform the installation, launch the server and help with connections to the main infrastructure.

As a result, the customer will not have to contact various firms, renegotiate contracts, pay for many additional services at all stages of work. It is enough to find your provider, who will do everything on a turnkey and one stop basis, as well as take care of all the paperwork.

When corporate data centers, high-performance computing companies and clouds need high-density servers, they choose blades. They are compact and can place, for example, 100+ servers in a rack for 42 units and more than 4000 cores.

This decision drastically reduces the cost of rent. You can buy a 24 or 48-core processor for each blade and get a super-performance computing system with 2,400 or 4,800 cores in a 42-unit rack. But at the same time, the disk system of blades is slower, so huge external data storages are connected to them.

When buying a server, it is better to estimate future costs (cost of ownership) in advance. For example, you can buy a 4-unit server for $100,000. If we take into account that the average cost of location in Europe starts from 100 euros per unit including electricity costs, then the rental fee will be up to 400 euros per month.

The server operation period is about five years. The total rent for this period is about 24 thousand euros. Suppose that when buying a server, errors were made in the configuration calculations and the server turned out to be bigger than necessary. If a higher-performance 1U server similar in price had been purchased initially, then variable costs would have been reduced by €18,000 over the same period. The ideal optimisation option is one-time capital expenditures with a noticeable decrease in variables. Therefore, in some situations it is easier to optimise your equipment and save on rent in the future.

The best time to agree on further cooperation with the equipment supplier or provider is only when all the preparatory calculations are carried out. As a rule, companies prefer to buy a server after calculations made with the configurator or custom calculation by the service provider (which is better). The cost of purchase depends on the amounts of components: case, processor, RAM, HDD, controller. The cheapest configuration is standard one, but it does not always fit the company’s tasks.

If the company has its own servers, additional hardware is often purchased to increase power to take advantage of the collocation feature. Some customers buy physical servers for virtual servers. The selection of such equipment is almost always a custom project.

On average, server equipment lasts for 3-5 years, then it requires replacement or modernisation, but not everyone considers it rational. Therefore, it is important for companies that the data center takes on the maintenance of old servers.

It is a common case when used equipment is purchased, sometimes with expired warranty. As a rule, the margin of safety is enough. But the older it is, the fewer providers are willing to take on such a server.

We have analysed 20+ European providers and made a checklist to help you avoid common pitfalls and choose a decent service provider even in complicated cases like discontinued equipment.

Having decided on the purchase, you can order a physical server both directly from the vendor and using the services of data centers (if the provider has already been chosen). If it is difficult to buy a new server due to budget limits, it is proposed to buy equipment in instalments. In this case, the data center usually offers a couple of additional years of free service.

The growth rate of business in the long run and benefits of the company from it depend on the service provider. To place a server, you have to search for a convenient data center that offers the most comfortable conditions and is located inside an Internet hub where the most of backbone communication channels converge. This speeds up the response time, increases the connection stability.

When choosing a data center, the customer primarily cares about guarantees of uninterrupted operation. A reliable data center provides for Tier 3 and 99.82% SLA availability. This means constant access to resources and elimination of failure consequences in the shortest possible time at the expense of the technical service of the contractor.

The physical server is the property of the customer, but at the same time its installation in the data center allows them not to bother with the arrangement of a dedicated server room in the office. Data centers provide:

• uninterrupted power supply (diesel-rotor UPS systems);
• climate control of machine rooms with equipment overheating/cooling control;
• remote hands to transfer server maintenance to the provider (optional, to be agreed).
• individual locks /authorization program providing technical personnel with access to the server for a limited period of time. The owner creates a temporary code to access the equipment, which is cancelled after the work is finished and the lock is latched.

To choose a data center, we suggest using the checklist prepared by our specialists based on the results of the analysis of the work of more than 20 data centers in Europe. Download the checklist for free.

Article author

Volodymyr Marchenko

network engineer and a consultant in interactions with leading equipment vendors and manufactures

Case study: Moving to a new data center in Germany and setting up a distributed infrastructure

READ MORE