Dedicated Server Selection: Specs, Bandwidth & Management

Dedicated Server Selection: Specs, Bandwidth & Management

We once migrated a UGC community with 200,000 daily active users: eight 4-core 8GB VPS instances could not handle the evening database connection spikes. After a month of sharding, cache tuning and scaling gymnastics, we moved to three 32-core 128GB dedicated servers. The evening peak stabilized and the monthly cost actually dropped by about 30%. That example is not an argument to jump straight to bare metal; it illustrates a judgment rule: when "shared" becomes the bottleneck, a dedicated server's "exclusive and predictable" profile earns its price.

The trade-off with VPS is covered in the VPS vs dedicated comparison, and with cloud in bare metal vs cloud. This guide assumes you have already decided to buy and focuses on how to choose.

1. Start with the CPU

The CPU sets the ceiling for what the machine can do:

Two reminders: first, a dedicated CPU usually cannot be swapped like RAM — upgrading means migrating the whole machine, so it is safer to order one tier higher for headroom. Second, for AI workloads pay attention to GPU memory and NVLink interconnects; see AI GPU bare metal.

2. Then Choose the Memory

• Prefer ECC memory: a server runs 24/7, and bit-flip "soft errors" are detected and corrected under ECC, whereas ordinary memory can silently write bad data.
• Capacity reference: 8GB to start; 16-32GB is common for PHP/Node sites; databases or in-memory caches (Redis) should go straight to 128GB+.
• Before ordering, confirm the number of motherboard slots and the max per-DIMM capacity, so you never buy a machine that cannot take more RAM later.

3. Storage and RAID

• System disks: 1-2 NVMe SSDs in RAID1 for double protection of the OS and data.
• Data disks: NVMe for hot access, HDD or storage-class models for large cold datasets.
• RAID choice: RAID1 mirrors the system and critical data; RAID5/6 use parity to balance capacity and fault tolerance (good for 3+ data disks); RAID10 balances performance and safety but doubles the disk count. RAID levels are detailed in server storage selection.
• At checkout, confirm three things: whether a hardware RAID controller is included (software RAID makes failed-disk recovery more painful), whether failed-disk replacement is billed, and whether the RMA process supports cross-region shipping.

4. Bandwidth and Network

Bandwidth is where dedicated servers bite the unwary:

• Port speed: 1G vs 10G sets the instantaneous throughput ceiling.
• Billing model: unmetered (usually with a capped port speed) vs metered traffic (pay for overage). Traffic-heavy businesses (downloads, video, APIs) must estimate monthly volume before choosing.
• DDoS protection: a dedicated server's public IP is almost guaranteed to be attacked at some point — confirm whether baseline protection is included and how scrubbing is billed.
• IP addresses: multi-domain, multi-certificate, or mail setups need to confirm how many IPs are included and the per-extra-IP price.
• Data center region: serve users nearby; decide with server region selection.

A concrete figure: a site with 10,000 daily users and 200KB average pages moves roughly 0.6-1.5TB a month, so a 1G unmetered port is usually enough. But video or download properties can push tens of terabytes a month — you will need metered billing and a clear picture of the overage rate.

5. Management and Support

• Unmanaged: the provider handles hardware and network only; you own the OS, security, and backups. Cheaper, for teams with ops skills.
• Managed: the provider handles OS-level maintenance (updates, monitoring, basic hardening). More expensive, far less hassle.
• Remote management: confirm IPMI/KVM is included so you can reinstall the OS, check hardware health, and power-cycle remotely — without it, a data-center incident means waiting for a human.
• SLA and hardware replacement: response and replacement times (e.g. 1-4 hours) should be written into the contract; but backups are always your own responsibility — see website backup strategy.

6. Cost Calculation

Dedicated servers are usually billed monthly, plus a one-time setup fee and possible bandwidth overages. Do not compare monthly fees alone — compare the three-year total cost of ownership against an equivalent cloud setup. Example: a 32-core 128GB dedicated server at ¥3,000/month costs about ¥108,000 over three years; an equivalent always-on cloud VM at list price can exceed ¥200,000. But if your demand only peaks for three months a year, cloud on-demand plus scaling wins instead. Estimate with the server cost calculator and cloud server pricing 2026, and see server operations tips for day-to-day work.

7. Sample Configuration Sheets

Business shapes differ hugely, so here are two reference configs:

Config A: High-concurrency web / small e-commerce (tens of thousands of DAU)

Config B: Data analysis / rendering / AI inference

Configs only answer "what to buy"; the earlier checklists still cover what to verify before ordering. And if your budget is elastic and you need capacity fast, running on cloud VMs first and migrating to bare metal later is a common path.

8. Decision Checklist

• CPU cores and clock cover peak load, with room to grow
• ECC memory, with slots that support expansion
• RAID plan clear, and failed-disk handling and fees understood
• Bandwidth billing, DDoS protection, and IP count confirmed
• Management tier and hardware replacement SLA in the contract
• Three-year TCO compared with cloud, with a documented conclusion

9. Common Questions

Q: When should I choose a dedicated server over a cloud VM? Choose dedicated when the workload is always-on, the specs are stable, and long-term cost-effectiveness matters; choose cloud when demand fluctuates and you need second-level scaling. One is built to "run long," the other to "run flexibly" — neither is inherently better.

Q: Do I really need RAID10? It depends on how critical the data is. RAID1 on the system disks is the bare minimum; on data disks, RAID10 balances performance and safety while RAID5/6 favors capacity per dollar — and in every case pair it with off-site backups.

Q: What if my dedicated server gets DDoS'd? Rely on the provider's baseline protection first, then hide the origin IP and put a high-capacity anti-DDoS CDN in front, and never let the real IP leak through DNS records.

Reference: https://www.ovhcloud.com/en/bare-metal/ , https://www.hetzner.com/dedicated-rootserver , https://en.wikipedia.org/wiki/Standard_RAID_levels