8 min readPacketStream
From what a VPS is and how it differs from shared hosting to choosing CPU, memory and storage, traffic billing and backups. The criteria you need when choosing a virtual server for the first time.
On this page
Even when you find two virtual server plans with the same vCPU and memory on a pricing table, it is hard to tell from price alone which one fits. The conditions for using the CPU, the included data transfer and the scope of backups can differ, and the work of managing the operating system still remains. When choosing your first VPS, the starting point is the programs you need and how you will operate them.
A VPS is a virtual server with its own operating system
A VPS (Virtual Private Server) is an environment that divides the resources of a physical server through virtualization. Each virtual server uses its own operating system and its allocated CPU, memory and storage. Within the administrative privileges allowed, the user can install programs and change service settings.
It is easier to understand if you picture a self-contained office unit inside a building. What is inside the unit is managed separately, but you do not have the building's infrastructure to yourself. A real VPS can also share physical hardware with other virtual servers, so separation of the operating system and exclusive use of physical resources need to be seen as two different things.
A VPS and a cloud VM are sometimes provided on the same virtualization technology. Rather than judging auto scaling or failover from the name, it is more accurate to compare the features actually supported and the pricing structure.
When to move from shared hosting to a VPS
A VPS becomes worth considering when you need direct control over the programs you install or over operating system settings. If you only run a website in a fixed environment, shared hosting can be enough. If you need separate background jobs, your own API or a specific runtime version, it is better to check first what range of settings the service allows.
| Criterion | Shared hosting | VPS | Dedicated physical server |
|---|---|---|---|
| Environment provided | Mainly a web runtime environment set by the provider | An independent virtual operating system | Sole use of a physical server |
| Scope of settings | Settings the plan allows | Settings subject to administrator privileges and plan policy | Settings subject to the hardware and hosting contract |
| Resource terms | Shared resources and per-account limits | Check the allocation and the CPU sharing policy | Check the server specifications and connectivity terms |
| Operational responsibility | Managed by the provider within its support scope | Check the scope of OS and application management separately | Depends on whether the contract is managed |
Moving to a VPS lets you do more, but security patches, log management and recovery procedures come into consideration with it. If the cause of a slow service lies in database queries or an external API, changing the type of server alone may not solve it.
The nature of the workload comes before the vCPU count
CPU, memory and storage each do a different job, so telling apart which resource is short reduces unnecessary upgrades. This is also why the right specification is hard to set from visitor numbers alone: the work handled differs with each request. Serving static files and an API with many database lookups use resources differently even at the same number of requests.
| Resource | Terms to check | Metrics to watch in use |
|---|---|---|
| CPU | Shared or dedicated policy, CPU generation, whether work can run in parallel | CPU utilization, run queue, processing time |
| Memory | Space the OS, applications and database will use together | Available memory, swap, out-of-memory kills |
| Storage | Capacity, IOPS and throughput limits, terms for expanding disks | Disk usage, I/O wait, read and write latency |
| Network | Port speed, included data transfer, the path to your users | Data transferred, latency, packet loss |
A program whose work piles onto a single thread of execution may not get that much faster when you add vCPUs. You can start by checking whether several tasks can run at the same time and whether one particular process keeps using the CPU. Memory is also used for the operating system's cache, so it is better not to conclude you are short from the “in use” percentage alone.
With disks too, the NVMe name and the performance actually provided need to be told apart. A workload that reads small files often and one that transfers large files sequentially require different performance. Comparing the limits the provider publishes with the results of your own workload makes it easier to decide which resource to increase.
Web, API and database
- Separate the response times of ordinary requests and slow requests.
- Check whether the application and the database share the memory of the same server.
- Tell database connection waits and disk latency apart from a CPU shortage.
Batch, build and development environments
- Look at CPU needs by the number of jobs that will run at the same time.
- Check the point where memory spikes when a job starts.
- Include the storage that build outputs, caches and logs will take up.
You do not need to choose the largest plan from the start. Starting from a configuration that meets the program's minimum requirements and checking a load similar to real requests in an environment you manage leads to a better-grounded choice. Before testing, though, check what the provider permits and its load testing policy.
Port speed and monthly data transfer are different terms
Port speed is the upper limit on the speed at which data is sent, and data transfer is the amount of data sent and received over a given period. Even with a fast port attached, traffic beyond the included allowance may be charged separately or limited. It is important not to compare the Gbps and the GB or TB on a pricing table as if they were the same measure.
Before buying, you can check whether outbound or inbound traffic is billed, whether the allowance is monthly or daily, and whether an overage brings extra charges or a speed limit. Backup file transfers and OS image downloads can also count toward usage depending on the policy, so it is safer to include the data movement that real operation will generate in your calculation.
Latency between your users and the server is a separate matter as well. Choosing a nearby region is the starting point, but the actual path is affected by the carrier in use and the state of the connection. For a service aimed at users in Korea, checking response and loss from several carriers helps isolate problems that occur on only one line.
The per-carrier measurement in PacketStream Pulse can be a reference when checking the path before and after you choose a server.
What the monthly fee includes and what the operator handles
The base server fee is one part of the total operating cost. Checking for separately billed items such as backup storage, additional disks, traffic overage and operating system licenses reduces the gap between the expected cost and the actual invoice. Whether managed services are included is also a term to compare along with price.
On a typical unmanaged VPS, the provider operates the physical infrastructure, but the user handles OS updates and application security. Even where a plan says it is managed, you can check the scope of the contract for what is managed on your behalf. “Supported” and “done automatically” are different promises.
Even where basic DDoS mitigation is provided, it does not resolve every web application vulnerability or wrongly opened management port. Allowing only the access you need and managing credentials and updates continues after the server is created.
A backup matters only when you can restore it
Backup frequency and retention period can be set by the data loss and downtime the service can bear. The fact that files were saved does not tell you whether the application will recover properly. You need to check that the database and the files are from matching points in time, and that settings and credentials can be recovered too.
A snapshot is a way to preserve the state at a point in time, but that does not guarantee it is a backup stored in a separate failure domain. You can check where it is stored, the deletion policy and whether recovery is possible when the original is damaged. It is safer to run restore tests in an environment separate from the production service.
- Recovery targetsDecide the range of data loss and the service downtime you will accept.
- Retention termsCheck the backup frequency, retention period, storage location and deletion permissions.
- Restore testRestore data and settings in a separate environment and confirm the application works.
The order for choosing your first VPS
A first configuration is a choice that fits the program's requirements, the location of your users and operational responsibility together. If you have no measurements yet, it is better not to treat estimates as a fixed specification and to leave them as items to confirm in testing before the service goes public.
- Runtime environment: Check the operating system and program versions, and the settings that need administrator privileges.
- Initial resources: Choose CPU, memory and disk based on the minimum requirements and the jobs that will run at the same time.
- Network and cost: Compare the location of your users, how data transfer is provided and the items billed extra.
- Operations and recovery: Decide who is in charge of updates and the backup and restore procedure.
- Terms for change: Check whether an upgrade requires a restart, the limits on shrinking disks and the procedure for moving to a different configuration.
PacketStream's Virtual Server provides KVM-based virtualization, SSH key authentication and VNC console access, firewall rules and basic DDoS mitigation. The specific configurations and pricing are on the product page. Adjusting the first configuration you picked from the spec sheet against real operating metrics is the final step of the choice.
Related products
More insights
All insights
DDoS Mitigation: How to Measure Attacks and What to Ask Before You Choose
DDoS mitigation means telling attack traffic from legitimate traffic. We cover L3/L4 and L7 attacks, bps, pps and rps, Always-On vs On-Demand, WAFs, questions to ask first and the order of response.Perspectives6 min read

Threat Intelligence: How Global Attack Data Becomes Mitigation Policy
A blocklist is just the output. From deep packet inspection of traffic across PacketStream's infrastructure to AI curation, verdict and scoring, policy distribution, and automatic enforcement in firewalls, CDNs and DDoS mitigation, here's the five-stage pipeline that turns attack data into real mitigation policy.Perspectives4 min read

L7 DDoS Protection: Why Application-Layer Defense Is Now Essential
DDoS attacks have evolved beyond the network layer to target application behavior itself. We examine the structural limits of L3/L4 defense and PacketStream's stateful, payload-based L7 protection.Perspectives5 min read
Ready to get started?
Tell us about your workload and we will help you choose the right products and plans.
Talk to our team
Questions about anything you read here? Our team will help.
Contact SalesInsights
Where PacketStream is heading, what we learn running the network, and the latest product news and notices.
All insights
