Understanding VPS Specifications: Memory, Storage, Traffic, and Bandwidth

选购 Cloud servers 或 VPS 时，配置列表通常由一串核心指标构成：vCPU 核心数、 Memory 容量、磁盘 Specification 、 Port speed 与每 Monthly transfer 。这些数字直接决定了服务器的承载能力、 Networking 体验和总体成本。然而，许多初学者往往仅根据数字大小或month付单价做判断，忽视了参数背后的计量单位、 resources 共享机制、流量统计规则以及控制面板的运维约束，最终在遇到性能卡顿、突发账单或服务停机时措手不及。

深入理解 VPS 的基础 Specification ，需要拆解 compute 、 Storage 与 Networking 三大子系统，结合真实 Providers （如 BandwagonHost 与 DMIT）的技术实现与运维规则，建立清晰的参数评估模型。

1. Compute Resources: vCPU/RAM Ratios and Overage Pitfalls#

Compute Resources Determine How Many Concurrent Tasks the Operating System and Applications Can Handle.

                    ┌────────────────────────┐
                    │       宿主机物理 CPU    │
                    └───────────┬────────────┘
                                │ 虚拟化划分 (KVM)
        ┌───────────────────────┴───────────────────────┐
        ▼                                               ▼
┌───────────────┐                               ┌───────────────┐
│  共享型 vCPU  │ ◄── 严格限制长期平均利用率    │  独享型 vCPU  │ ◄── 允许长期 100% 满载
│ (10%~30% 限制) │     超时触发降频或强制休眠    │ (High-CPU/Pro)│     价格较高，面向生产
└───────────────┘                               └───────────────┘

vCPU: Shared Scheduling and Credit Mechanisms Behind the Core Count#

The “1 vCPU” or “2 vCPU” listed in a plan usually represents virtualized thread slices from a physical CPU's Hyper-Threading capabilities, commonly single-thread mappings under KVM. Most entry-level and standard VPS plans fall into共享型虚拟核心。

• CPU Contention and Steal Time: Multiple tenants share the same host. When neighboring tenants simultaneously demand compute resources, your virtual machine may have to wait for scheduling even with a heavy workload. On Linux, run top command, you can see %st(Steal Time) metric. If it stays above 5%~10%, the host is experiencing clear contention for computing resources.
• Fair Usage Guidelines (AUP): major providers impose strict usage thresholds on shared vCPUs. Keeping 1 shared CPU core at 100% for extended periods, such as continuously compiling source code, running complex models for long periods, or executing an infinite loop, triggers hard Throttling by the host scheduler or even risk-control suspension. If your workload requires uninterrupted heavy computation, purchase a Dedicated CPU plan.

Memory (RAM): Physical Capacity and Swap Space#

Memory 是程序执行过程中的工作区。不同技术栈对物理 Memory 的需求存在质的差异：

• Baseline Usage and System Overhead: A newly installed minimal Debian or Ubuntu system, with no additional services, typically uses about 80 MB~200 MB of memory at idle after boot.
• Application BaselineLightweight web services, such as Nginx/Caddy + SQLite + a static site, can run well on a server with 512 MB of RAM. But adding a MySQL/MariaDB relational database, Docker engine, Java runtime, or PHP-FPM multiprocess pool makes 512 MB highly prone to triggering the Linux kernel's OOM Killer (Forced Process Termination When Out of Memory), causing the database service to crash without warning.
• What Swap Really Is: When physical memory is exhausted, the system writes cold data to disk-based virtual memory (Swap). On NVMe/SSD storage, Swap can prevent an immediate out-of-memory crash, but it is several orders of magnitude slower than physical RAM. Frequent Swap I/O directly causes storage congestion and high CPU softirq usage. When evaluating plans, physical RAM must cover the core workload's steady-state peak; Swap cannot serve as the main memory.

2. Storage 空间：标称容量、实际可用性与 I/O 瓶颈#

Storage is not only about how many GB of data you can keep, but also about read/write throughput and data security.

Conversion Discrepancies and Actual System Usage#

After purchasing a VPS with a 20 GB disk, log in and run df -h , you may find that the root partition totals only about 18.5 GiB. This is not the provider shortchanging you; it is caused by two factors:

1. 十进制与二进制的单位差异: Hardware manufacturers and plan specifications usually use decimal units (1 GB = $10^9$ Bytes), while operating system filesystems use binary units (1 GiB = $2^{30}$ Bytes $\approx$ 1.074 GB). Thus, 20 GB is approximately 18.62 GiB in the operating system's units.
2. System Overhead and Reserved Blocks: Linux ext4 filesystems reserve 5% of disk space for emergency root maintenance by default. In addition, a Linux distribution's base system image and package management database typically occupy 1.5 GiB3 GiB. With a 20 GB plan, the space actually available for applications and data is usually between 14 GiBand 16 GiB.

Storage Protocols and I/O Performance#

• Differences in Storage Media: SSD storage is now standard across VPS offerings. Those labeled NVMe SSD models connect directly over PCIe, delivering 4K random read/write IOPS and sequential throughput far beyond older SATA SSDs.
• Shared Host I/O Throttling: to prevent a single user's highly concurrent disk activity from overwhelming the entire host's storage array, providers generally impose I/O limits at the virtualization layer (such as a maximum of 500 MB/s or 10000 IOPS).

Disk Lifespan and Data Retention#

Providers differ greatly in how they handle nonpayment suspensions, refunds, and data deletion. These policies directly determine whether an instance's disk survives:

• Data Retention Period After Expiration Without Renewal: Using DMIT 为例，其 instance month付账单通常在到期前 7 days生成；当 instance 到达到期日而未支付续费时，机器会进入暂停（Suspended）状态。官方规则明确该状态一般仅保留 3 days, if the invoice remains unpaid beyond the deadline, the system automatically destroys the underlying storage volume and irreversibly erases the data.
• Refunds and Immediate Destruction: in BandwagonHost or DMIT, once the refund is confirmed and processed, the corresponding instance and virtual disk are immediately reclaimed and erased. Keeping the only copy of your data on a VPS disk is extremely risky; production services require regular external object-storage backups or remote synchronization.

3. Bandwidth and Ports: Speed Conversion, Shared Bursts, and Emergency Management#

Network transfer performance depends on both port speed and route quality; the two should not be treated as equivalent.

Converting Between Mbps and MB/s#

Bandwidth is usually measured in bit/s (Bits per Second) as the unit, while download tools and system monitoring panels usually use Byte/s (Bytes per Second) shows:

$\text{Actual Theoretical Download Speed (MB/s)} = \frac{\text{Port Bandwidth (Mbps)}}{8}$

• 1 Gbps Port: The theoretical maximum is $1000 \div 8 = 125 \text{ MB/s}$. After accounting for TCP/IP header overhead (MTU/MSS), retransmissions, and congestion-window fluctuations, steady-state public-network peak transfers are typically around 100 MB/s~115 MB/s.
• 100 Mbps Port: the theoretical maximum is approximately $12.5 \text{ MB/s}$, with a sustained peak of about 11 MB/s.

共享峰值（Burst）与 Networking 路由权衡#

Backbone bandwidth procurement costs are extremely high for most premium international routes, including China Telecom CN2 GIA, China Unicom 9929/AS4837, and China Mobile CMIN2 optimized networks.

• On these expensive routes, providers generally offer High-bandwidth ports (such as shared peak speeds of 1 Gbps~10 Gbps) paired with a capped monthly traffic allowance policy. This means you can briefly burst to download large files during traffic peaks, but it does not mean you can saturate a gigabit connection for 24 hours.
• Standard unoptimized international routes, such as Hurricane Electric and ordinary Cogent transit, are inexpensive and may offer larger allowances, but cross-border, cross-carrier access can suffer greater packet loss and latency variation.

              ┌──────────────────┐
              │   公网 SSH 连接   │
              └────────┬─────────┘
                       │ 网络失联 (防火墙误配/路由中断)
                       ▼
┌──────────────────────────────────────────────┐
│            服务商控制面板应急通道            │
├──────────────────────┬───────────────────────┤
│ BandwagonHost        │ DMIT                  │
│ KiwiVM Interactive   │ Web Serial Console    │
│ Console              │ (网页直接打开文本终端)│
└──────────────────────┴───────────────────────┘
                       │ 绕过公网网络层
                       ▼
            ┌──────────────────────┐
            │   VPS 实例内核交互   │
            └──────────────────────┘

Lost Network Access and the Emergency Console#

当我们在 VPS 内部误配了 iptables/ufw firewall, a disabled SSH port, or routing instability prevents remote SSH connections, many people mistakenly assume the server has crashed completely.

• What the Console Really Is: in BandwagonHost provides the following in its KiwiVM panel: Interactive Console，而在 DMIT provides the following on its instance management page: Console feature.
• Rescue Mechanism: These consoles bypass the VPS's public network interface and connect directly to the virtual machine's terminal output through underlying virtualization interfaces such as VNC or a Serial Console. Even if all server networking is blocked, as long as the operating system is running, administrators can log in through the panel Console to inspect logs, repair firewall rules, or bring network interfaces back up.

4. Monthly Traffic and Billing: One-Way versus Two-Way Metering, Overage Policies, and Refund Limits#

Monthly Data Transfer allowances are the main way providers limit network consumption. The same advertised 1000 GB per month can yield twice as much usable traffic under one provider's accounting method as another's.

What Happens After the Traffic Quota Is Exceeded#

When you exhaust your traffic allowance within a billing period, providers generally use one of two approaches:

1. Automatic Suspension (Suspended / Hard Cap): The network port is immediately disconnected and public services are suspended until the next billing cycle resets the allowance, unless the user purchases an additional traffic package.
2. Bandwidth Throttling (Throttled / Soft Cap): The instance remains online, but its public port speed is reduced to 1 Mbps~10 Mbps, allowing only very slow maintenance access. Before purchasing a specific plan, check its traffic policy and confirm whether the allowance resets on the first day of the calendar month or according to the invoice date (Billing Cycle).

流量消耗与 Providers 退款门槛#

Providers commonly use traffic consumption as a strict refund eligibility limit to prevent abusive testing or temporary exploitation of high-bandwidth resources:

• Data Usage Limits in BandwagonHost's Refund Policy: under its Terms of Service (ToS), eligible newly purchased services may receive a full refund within 30 days, but one prerequisite isMonthly Traffic Usage Must Be Strictly Below 10% of the Total Allowance. If you purchase a plan with 1000 GB of traffic and use 150 GB, you lose refund eligibility because the traffic-percentage limit has been exceeded, even within the 30-day window.
• Absolute Traffic Limits in DMIT's Refund Policy：根据 DMIT Docs 的退款 FAQ，符合条件的新购 instance 申请 3 days内全额退款时，明确要求Actual Network Usage Must Not Exceed 30 GB; after 3 days but within the 30-day period, a partial refund is calculated based on the remaining value, subject to the same strict traffic-quota usage ratios and compliance terms.

5. Network 与鉴权：IP 状态、SSH 密钥与控制面板层级#

A VPS's network entry point is its IP address, its security barrier is credential management, and its highest administrative authority resides in the provider's control panel.

                    ┌────────────────────────────┐
                    │      服务商客户中心账户     │
                    │ (账单支付 / 续费 / 工单)   │
                    └─────────────┬──────────────┘
                                  │ 独立的鉴权层级
                    ┌─────────────┴──────────────┐
                    │     主机管理面板控制台      │
                    │ (KiwiVM / DMIT Instance UI)│
                    └─────────────┬──────────────┘
                                  │ 独立的底层管理与救援
        ┌─────────────────────────┴─────────────────────────┐
        ▼                                                   ▼
┌───────────────────────────────┐   ┌───────────────────────────────┐
│     BandwagonHost 模式        │   │           DMIT 模式           │
├───────────────────────────────┤   ├───────────────────────────────┤
│ • KiwiVM 独立管理密码         │   │ • 默认禁用 root 远程密码登录  │
│ • 系统 root 密码由面板生成    │   │ • 强制/推荐配置 SSH 公钥      │
│ • IP 异常时机房迁移功能受限   │   │ • 面板更换密钥后必须重启生效  │
└───────────────────────────────┘   └───────────────────────────────┘

IPv4 / IPv6 Allocation and IP Status#

• Dedicated IPv4: the vast majority of standard VPS plans include one or more dedicated public IPv4 addresses. A dedicated IPv4 lets you freely listen on standard service ports (such as 80, 443, and 22).
• IP Reputation and Feature Restrictions: the quality of a VPS's public IP directly affects network functionality. For example, in BandwagonHost the system offers the useful ability to migrate freely between data centers (Migrate to another DC). However, according to its official knowledge base,If the instance's current IP is on a network blacklist, such as a GFW block or an Abuse-related ban, the panel restricts data center migration. Resolve the IP status issue first before migration access can be restored.

Remote Login Security: Passwords and SSH Keys#

A Linux server's port 22 exposed to the internet faces thousands of automated brute-force attempts daily. Major providers use different security strategies during initialization:

• DMIT's Default Security Baseline: In DMIT's newer images,Remote root Password Login Disabled by Default。系统推荐并鼓励用户在下单及 instance 管理界面中使用 SSH Keys. Users can upload public keys to the SSH Keys repository in their DMIT account and directly associate or replace them through the instance's Access menu.需要注意的是: according to the official documentation, after updating the key in the panel,The Instance Must Be Restarted Through a Panel Command, so the virtual machine's initialization process can correctly inject and activate the new key.
• BandwagonHost's Multi-Layer Password Isolation: BandwagonHost strictly separates three completely independent password levels:
  1. Client Area Password: Used to log in to billing, submit tickets, and make payments;
  2. KiwiVM 管理面板密码: Used to manage underlying power controls, mount ISOs, and migrate data centers;
  3. Operating System root Password: Superadministrator credentials for the virtual machine's internal operating system. The three credentials are independent. After OS reinstallation, KiwiVM generates and displays a new random root password, strengthening security isolation between account access and server operations.

6. 参数如何落地：账单周期与运维排查清单#

When reading specifications and preparing to deploy real workloads, evaluate static resources (CPU/RAM/Disk/Bandwidth) together with the ongoing operational lifecycle (renewal/troubleshooting/backups).

Renewal Mechanisms and Payment Risks#

• Automatic Payments and Invoice Generation Cycles：
  • BandwagonHost 的续费 knowledge base 明确指出：系统Will Not从绑定的信用卡或 PayPal 账户中发, from未经授权的自动扣款（Auto-charge）。如果该服务的续费处于开启状态，系统会在到期前 7 days 自动生成续费账单；此时只有当你的账户预先充值了足够的余额（Account Balance）时，系统才会自动划扣余额完成续费，否则必须由人工手动前往账单页面完成支付。
  • DMIT monthly renewal invoice is likewise issued before expiration 7 days 生成，若在到期后未能及时续费，机器停机后只有短暂的 3 days宽限期。
• Planning Strategy: For critical production workloads, set recurring calendar reminders or add sufficient account credit in advance to prevent essential data from being destroyed after the grace period because an email notice was missed.

Comprehensive VPS Specification Quick Reference#

When choosing and configuring a cloud server, understanding hardware limits, billing measurement rules, and the provider panel's underlying emergency capabilities is essential to reliable operation and effective cost control.