Knowledge about server bandwidth and traffic sizing
Bandwidth is one of the most common pitfalls when launching a website or web application. If you book a VPS and get throttled or even start serving 503 errors on your first viral tweet, you don't lack power, you lack egress bandwidth. The KernelHost bandwidth calculator helps you estimate realistic monthly traffic demand and pick a matching tier before you see your first invoice.
The math behind it is trivial. You multiply the number of daily visits by the average page weight in MB and by 30 (days per month). On top you add an egress overhead because HTTP headers, TLS handshakes, API calls and the occasional retransmit produce extra traffic that isn't part of the raw page weight. Divided by 1024 you get gigabytes, divided by another 1024 terabytes. Three inputs, one reliable sizing result.
What many people underestimate is that modern websites are far heavier than they were five years ago. The HTTP Archive Report (as of 2026) puts the average page weight at around 2.5 MB on desktop and 2.2 MB on mobile. That covers HTML, compiled JavaScript, CSS, fonts and images of a typical landing page. Media heavy sites with hero videos, high resolution sliders or embedded WebGL easily climb to 5 to 15 MB. People running an unoptimised WordPress theme often hit 4 to 6 MB per page without noticing.
When entering values, look at the real numbers. Tools like WebPageTest, Lighthouse or your browser's network tab show you the exact transfer size of your page (after Brotli compression). Use that as page weight, the calculator adds the overhead automatically for a realistic picture.
How to calculate server bandwidth correctly?
The base formula is: daily visits * page weight in MB * 30 * (1 + overhead/100) / 1024 = monthly bandwidth in GB. Concrete example: 5000 daily visits at 2.5 MB page weight and 20 percent overhead come out at 5000 * 2.5 * 30 * 1.2 / 1024 = 439 GB per month, or roughly 0.44 TB. A KernelHost VPS S with 1 Gbit unmetered would have plenty of headroom here.
Always size with reserve. Cutting it close means you have an immediate problem during a viral spike or a short marketing push. Rule of thumb: plan for double the calculated value as headroom so you absorb burst traffic, growing customer counts and the occasional crawler burst. A 1 Gbit uplink can theoretically push 324 TB per month over a long average, in reality the limit lands closer to 30 to 50 TB on the 95th percentile model.
Second aspect: burst traffic vs. sustained traffic. Most hosters and carriers bill on the 95th percentile model. They sample throughput every 5 minutes, sort the values across the month, drop the top 5 percent and bill the highest remaining value. Short spikes (e.g. a 30 minute burst from a Hacker News article) don't show up in your bill at all. KernelHost VPS tiers are unmetered within their uplink speed, which sidesteps this discussion for most customers.
What is egress overhead?
Egress overhead is the extra traffic produced by the HTTP/HTTPS protocol mechanics that doesn't show up in raw asset volume. This includes TLS handshake data (about 5 to 10 KB per new connection), HTTP headers (cookies, cache control, auth tokens, on average 1 to 4 KB per request), TCP retransmits on flaky connections, JSON API calls for search, filtering and personalisation, plus tracking pixels and third party scripts.
On a typical website this adds up to 15 to 25 percent on top of the page weight. SaaS applications with many API calls and real time features (WebSockets, server sent events) sit closer to 30 to 50 percent. Streaming platforms with ABR (adaptive bitrate) sit around 5 to 10 percent due to constant manifest polling. The default of 20 percent is a conservative estimate for an average site with normal cookie load and a handful of third party trackers.
If you want exact numbers, compare your monthly bandwidth values from Cloudflare Analytics or your server access log against the theoretical value from page views times page weight. The delta is your real overhead in percent. On KernelHost servers you find the numbers in nload, vnstat or Netdata, alternatively in the customer portal per VLAN.
VPS vs. dedicated server at high bandwidth
For moderate traffic up to about 5 TB per month, a KVM VPS is usually the most economical pick. You share physical hardware with other tenants but get dedicated RAM, dedicated vCPU cores and a dedicated 1 Gbit uplink. KernelHost VPS tiers all run on NVMe SSD and 100 percent green power in Frankfurt FRA01, the same datacenter standard as the dedicated servers.
Above 5 to 10 TB egress the move to a dedicated server pays off. You get the full bandwidth of the physical uplink (1 or 10 Gbit, depending on the model), no noisy neighbour risk and full control over CPU, RAM and storage. For streaming platforms, CDN origins and database heavy SaaS apps, dedicated isn't just a performance question but a cost question, because high bandwidth VPS tiers get pricier than a comparable dedicated server.
Rule of thumb: up to 1 TB web hosting, up to 5 TB VPS S/M, up to 20 TB VPS L or Dedicated Entry, beyond that Dedicated Premium or Custom. These thresholds aren't hard, they're a consequence of typical hardware utilisation. If you have heavy burst requirements (live streaming events, software releases), a 10 Gbit dedicated with anti DDoS filtering in front is worth it because spikes don't trigger throttling.
When does CDN offload pay off?
A CDN (content delivery network) like Cloudflare, Fastly or BunnyCDN caches static assets (images, CSS, JS, fonts, videos) on edge nodes worldwide. When a visitor opens your page, the browser fetches the HTML body from your KernelHost server but pulls all other assets from the closest edge node. Origin traffic drops dramatically, typically by 60 to 90 percent depending on your cache configuration.
CDN starts to make sense from around 500 GB origin traffic per month or for international audiences. Cloudflare's free tier offers unlimited bandwidth, so it costs nothing but already brings 70 to 80 percent offload. Fastly and BunnyCDN have paid tiers from 0.005 USD per GB egress, which works out to roughly 50 USD per month at 10 TB but delivers 95+ percent offload. A self hosted cache layer with Varnish or nginx cache on a second KernelHost VPS is an alternative if you want to avoid third party dependencies.
Important: dynamic traffic (user specific HTML, REST API calls, authentication) still hits your origin. People serving 90 percent static markup see massive savings. People running a SaaS app with personalised responses on every request benefit less. The CDN toggle in the calculator assumes 70 percent offload, which is a realistic average for most sites.
DDoS protection and bandwidth: why 3.2 Tbps Arbor matters
When sizing bandwidth, people often overlook that not only legitimate traffic loads your uplink. A DDoS attack can saturate your 1 Gbit interface within minutes, so legitimate visitors get 503 errors and the server is effectively offline even though it sits idle on CPU. Most volumetric attacks (UDP floods, reflection amps via NTP, DNS, memcached) currently land between 100 Gbps and 1 Tbps, the largest record attacks (Cloudflare Q1 2024) hit 5.6 Tbps.
KernelHost runs a 3.2 Tbps Arbor DDoS mitigation system in Frankfurt FRA01, active on every customer VLAN interface. Included with every VPS and dedicated, no extra cost. The system scrubs malicious traffic at layer 3 and 4 (SYN floods, UDP amps, reflection attacks) before it ever reaches your server. Layer 7 attacks (HTTP floods, slowloris) are caught with nginx rate limiting or an upstream Cloudflare Pro setup if you opt for it.
In practice this means: your calculated bandwidth demand is the load for legitimate traffic. Volumetric attacks don't eat into your uplink budget, because they're scrubbed in front of your VLAN. The KernelHost infrastructure in Frankfurt FRA01 (Maincubes, Tier III, BSI IT Grundschutz compliant) keeps your server reachable even under attack, which no hyperscaler delivers at this price tier.