Clusters servers
| Component / Term | Description |
|---|---|
| Node Server | An individual server within a cluster that performs calculations and tasks. |
| Rack Server | A standardized structure that hosts multiple servers, with access to cooling and power. |
| Cluster Server | A group of servers connected to work together, providing scalability and high availability. |
| Server | A dedicated computer used for applications, databases, or other services. |
| Storage Units (NAS/SAN) | Solutions for sharing storage capacity among the servers in a cluster. |
| Network Switch | A device that connects servers in a high-speed network for efficient communication. |
| UPS (Uninterruptible Power Supply) | Uninterruptible power supply system, for protection against power outages. |
| Orchestration Software | Software like Kubernetes used to distribute tasks and scale resources. |
| Cooling Systems | Air or liquid cooling to keep servers at optimal temperatures. ( Termal Fluid 3M Novec ) |
| Centralized Management | BMC/IPMI consoles for remote server management, monitoring, and troubleshooting. |
To understand and calculate TFLOPS and GT (Gigateraflops), we first need to understand what FLOPS is and how computing power is measured for a processor.
FLOP stands for "Floating Point Operation" - a mathematical operation involving floating point numbers. It is a unit of performance measurement that reflects the number of complex mathematical calculations a processor can perform in one second. FLOPS stands for "Floating Point Operations Per Second," which means the number of such operations performed in one second.
- TFLOPS (teraflops) means one trillion (10^12) floating point operations per second.
- GT (gigateraflops) represents one million teraflops or 10^18 floating point operations per second.
The computing power (in TFLOPS) of a processor depends on several factors, such as:
- Number of Cores: The number of physical cores in the processor. Each core can perform calculations independently.
- Clock Speed: The clock speed of each core, measured in gigahertz (GHz). This indicates how many operations a core can execute per second.
- Operations per Clock Cycle: The number of floating point operations each core can execute in a single clock cycle. Modern processors can execute multiple operations per cycle, which makes them more efficient.
The theoretical power in TFLOPS can be calculated using the following formula:
[ \text{Computing Power (FLOPS)} = \text{Number of Cores} \times \text{Clock Speed (GHz)} \times \text{Operations per Cycle} \times 10^9 ]
To get TFLOPS, divide the result by (10^{12}).
Consider an Intel Core i7 processor with the following specifications:
- Number of Cores: 8
- Clock Speed per Core: 3.5 GHz
- Operations per Cycle: 16 (this varies depending on the architecture, but we'll assume 16 for this example)
Using the formula above:
- Number of Cores = 8
- Clock Speed = 3.5 GHz = (3.5 \times 10^9) Hz
- Operations per Cycle = 16
[ \text{FLOPS} = 8 \times 3.5 \times 16 \times 10^9 = 448 \times 10^9 \text{ FLOPS} ]
To get TFLOPS:
[ \text{TFLOPS} = \frac{448 \times 10^9}{10^{12}} = 0.448 \text{ TFLOPS} ]
Thus, an Intel Core i7 processor in this example has approximately 0.448 TFLOPS.
If you have a cluster of multiple such processors, the total computing power is obtained by adding the TFLOPS of all processors in the cluster. For example, if you have 100 such Intel Core i7 processors:
[ \text{Total Power} = 100 \times 0.448 \text{ TFLOPS} = 44.8 \text{ TFLOPS} ]
- 1 GT (Gigateraflop) is equal to 1 million TFLOPS.
- If you consider the performance scale of a large data center capable of reaching 1 GT, it means they have a cluster of computers with a total power of 1 million TFLOPS or 10^18 FLOPS.