Web host performance in 2025: test of 10 offers

Accommodation providers are listed in alphabetical order, and rates are given annually and exclusive of tax.

HaiSoft

HaiSoft is a French hosting company with almost 25 years' experience. Its servers are located in Tours and Vélizy, France. HaiSoft offers managed virtual and dedicated servers, but this test covers two of their shared offerings.

For HaiSoft, we tested the following offers:

• HaiSoft Essential (€47.88 excl. VAT/year)
• HaiSoft Pro + Advanced WordPress Support (179.88€ HT/year) - Rated "HaiSoft Pro WP" during this test

Ionos

Ionos (formerly 1&1) is a German hosting provider. Ionos operates 31 data centers worldwide, including one in Paris and a second soon to be operational.

For Ionos, we tested an offer dedicated to WordPress :

• Ionos Expand (€12 excl. tax/year for 1 year, then €120 excl. tax/year)

LRob

LRob is an independent hosting company based in France. Servers are located in Germany in Hetzner datacenters. As a WordPress specialist, LRob offers optimized high-performance servers.

We have included our 4 single-site mutualized offers:

• LRob Starter Web (120€ HT/year) - Rated "LRob Starter" during this test
• LRob Pro Web (€190 excl. VAT/year) - Rated "LRob Pro" during this test
• LRob Super Web ( 300€ HT/year) - Rated "LRob Super" during this test
• LRob Ultra Web (480€ HT/year) - Rated "LRob Ultra" during this test

The tests were carried out under real-life conditions on the LRob servers used for the customers.

LWS

LWS is a French hosting provider that has been on the market for over 20 years, providing more than 680,000 web hostings. Its datacenters are located in France, and its front-end servers are renewed every two years to guarantee optimum performance.

For LWS, we tested a cPanel offer:

• LWS cPanel L (€59.88 excl. VAT/year for 1 year, then €83.88 excl. VAT/year)

OVH

OVH is a French hosting company with 1.6 million customers and 43 datacenters worldwide, including several in France. OVH emphasizes its performance/price ratio.

For OVH, we tested the second-price shared offer:

• OVH PERSO (€39.48 excl. VAT/year)

o2switch

o2switch is a popular French hosting provider with over 700,000 hosted sites. o2switch boasts "Monstrous Resources" and servers optimized for security and speed.

For o2switch, we chose their most affordable offer for the first year:

• o2switch Cloud. (€21.12 excl. tax/year then €192 excl. tax/year or €324 excl. tax/3 years)

Test methodology

Tested website

For this test, we created a very simple WordPress site based on the site's content. www.lrob.fr. The articles were imported using native WordPress tools.

• WordPress version : Latest version available at time of testing - WordPress 6.7.2.
• Indexation The site is configured not to be indexed by search engines in order to limit external traffic.
• Privacy URLs are not provided to help obtain an isolated test.

Installed WordPress plugins

The 6 to 7 plugins installed and active are as follows:

• Easy WP SMTP 2.10.0 - SMTP configured
• Go Live Update Urls 7.0.2 - Changing URLs in articles
• RankMath SEO 1.0.239 - Basic configuration
• W3 Total Cache 2.8.6 - Redis cache if available, otherwise "Disk Basic" (disabled for uncached tests)
• WordPress Importer 0.8.3 - Importing items
• WP Statistics 14.12.5 - Additional load generation
• WPForms Lite 1.9.4.1 - Additional load generation (default form)

Note: For Ionos, we deactivated the pre-installed plugins. The "Performance" plugin did not seem to have any impact in our test and was therefore not used for cached tests.

WordPress theme

The theme used is the native theme Twenty Twenty-Five (1.1).

The site comprises a single home page containing :

• Header Logo, site title, menu with one item, list of categories.
• Page body :
  • Large logo
  • Securities
  • List of 30 items with images
  • WPForms contact form (default form)
• Footer Logo, site title, site tagline, default menu items (8 items), theme name and link "Designed by WordPress.

Type of tests performed

Our tests focused mainly on response time of the WordPress site on each hosting. We carried out three separate series of tests:

• GTMetrix GTMetrix response test from the London (UK) server, with and without cache. This test measures several key performance metrics.
• ApacheBench Load test with ApacheBench 2.3, performed on an ARM Hetzner VPS in Nuremberg (Germany). This test measures the number of pages served per second under different conditions (cached and uncached).
• Uptime Kuma Measuring the average response time every 60 seconds over a 24-hour period, without caching, from the same ARM Hetzner VPS, which is an LRob monitoring server running Uptime Kuma 1.23.16.

Test details

GTMetrix : Loading speed tests

• The GTMetrix tests were carried out with a GTMetrix Micro offer from the London server, the closest to France.
• Results may be influenced by the quality of the network interconnection between the host and the London server.
• To guarantee representative measurements, we have carried out up to three tests per accommodation and selected the best TTFB (Time to First Byte) of the three.
• This procedure was repeated twice: once uncovered and once with cover.
• Times greater than 1000 ms are rounded to the nearest tenth of a second (example: 1.256 s ➔ 1.3 s).

ApacheBench Load test

• We initially tried to run these tests from an Orange Pro FTTH connection, but network limitations led us to use a Hetzner server we already had.
• The ApacheBench tests were performed from a VPS ARM Hetzner [uptime.lrob.net] 2vCores located in Nuremberg (Germany).
• We checked that the VPS was not saturated during the tests.
• Tests were carried out with a variable number of simultaneous connections to simulate different loads.
• Some hosters (LWS, o2switch, and Ionos) seemed to be applying a rate limiting (voluntary limitation of the number of requests from a single IP), which may have influenced the results.

Uptime Kuma

• The Uptime Kuma measurements were taken from a VPS ARM Hetzner [uptime.lrob.net] 2vCores located in Nuremberg (Germany), taking advantage of an already operational instance. This server is sometimes referred to as a "monitoring server" in the article.
• Uptime Kuma measurements were carried out every 60 seconds over a minimum period of 24 heures of which Uptime Kuma provides an average.
• Exception for LWS: a measurement every 60 seconds gave an average response time of 2,9 s while the real time seemed to be 900 ms after a first access. This difference could be the result of a resource-saving strategy on the part of the host. We therefore used a polling rate of 20 seconds instead of 60.
• We have supplied a weighted measured average values of 2x ping between [uptime.lrob.net] and the various hostings, to avoid the geographical advantage of certain servers over the monitoring server.
• Visit inaccessibility are also reported with the error messages returned by Uptime Kuma. These are highly variable over time, and are assumed to be unrepresentative given the short measurement period.

Cache configuration

For tests with cache, we have configured the plugin W3 Total Cache to activate only the page cache :

• Redis was used when available.
• Redis is an in-memory database, offering fast read speeds without saturating disk storage.
• LRob uses Redis by default on its customers' sites - so we've chosen this configuration as a reference.

Redis incompatibility :

• Offers OVH and Ionos didn't seem to support Redis in our configuration.
• We therefore used W3 Total Cache's "Disk Basic" cache instead.
• This difference may put these hosters at an advantage or disadvantage compared with those using Redis. These results should therefore be interpreted with caution.

Technical biases and external factors

• Visit peering (network interconnection) between Hetzner and the various hosting providers can impact results.
• In particular, as LRob is hosted by Hetzner, it could benefit from lower latency during testing.
• Network latency can vary from an order of 20 milliseconds between some hosts and Hetzner, which may influence the results.
• Some of the network outages observed could be attributed to the interconnection between Hetzner and French hosting providers.
• The dates and times of the tests are not particularly chosen. They simply correspond to the times when we were available to carry out the test.

Considerations on the domains used

• For the Ionos test, we used a sub-domain in .live-website.comsupplied by Ionos.
• The other tests were carried out on sub-domains in .bench.lrob.net pointing to the IPv4 and IPv6 addresses of the hosts concerned.
• This difference in DNS configuration could lead to slight variations in DNS resolution times and consequently in some response times.

Load testing with ApacheBench

These tests assess the maximum capacity of each accommodation manage a high volume of simultaneous requests using the ApacheBench (ab).

ab -l -c ${c} -n ${n} https://${url}/

ApacheBench load tests - No cache

This test measures the server's ability to dynamically generate pages under strong loadwithout the aid of a cover.

Test configuration

We have simulated the loading of 100 pages with the following configurations :

• 1, 2, 5, 10, 20 and 40 simultaneous connections
• The cache is disabled to measure raw server performance in terms of PHP/MySQL processing.

This test highlights the performance available when the server is either little usedor close to saturation.

Aggregate results - ApacheBench - No cache

The raw values of this benchmark are made available for download to anyone wishing to re-analyze them.

First, here's the general graph, followed by the table showing these values.

👉 For best performance, the aim is to get the highest possible bars. The aim is to compare each color with the next. Next, we detail the results obtained for each load level.

On this test, the LRob Ultra achieves the best results in all configurations. It is important to remember, however, that this is the most expensive of the offers tested.

1 connection - ApacheBench - No cache

On a single connection, servers offering TTFB lower (faster initial response time) are distinguished by higher loading speeds. The four offers LRob occupy the top spots.

Bids follow in the following order:

• LRob Starter, Pro, Super, Ultra - Between 4.21 and 4.25 pages/second
• Ionos Expand - 2.35 pages/second
• o2switch Cloud - 2.27 pages/second
• HaiSoft Pro WP - 2.12 pages/second
• OVH PERSO - 1.84 pages/second
• HaiSoft Essential - 1.71 pages/second
• LWS cPanel L - 1.1 pages/second

👉 In single connection, LRob loads between 1,8x and 3,86x more pages per second than the other offers tested.

2 connections - ApacheBench - No cache

On two simultaneous connectionsperformance is approximately lined for all hosters, which suggests that none of the rate-limiting has yet to be applied at this stage.

👉 The results observed are as follows:

• Offers LRob are in first place, with performances ranging from 8,07 and 8.15 pages/second.
• Offers from HaiSoft, Ionos, o2switch and OVH follow, with results ranging from 3,57 and 4.36 pages/second.
• The offer LWS cPanel L is slightly behind with 2.23 pages/second.

LRob maintains a similar lead in this test, with a capacity around 2x greater than the other offers in this test.

5 connections - ApacheBench - No cache

From 5 simultaneous connectionsThe performance differences between accommodations become more pronounced. Lodgings with 2 FPM or applying rate-limiting inevitably see their performance or readings drop.

👉 The results observed are as follows:

• Offers LRob Pro, LRob Super and LRob Ultra remain in the lead with performances between 17,88 and 19.29 pages/second.
• The offer o2switch Cloud ranks fourth with 10.77 pages/secondfollowed closely by HaiSoft Pro WP (9.82 pages/second) and LRob Starter (9.51 pages/second).
• OVH PERSO reaches its peak with 5.98 pages/secondslightly exceeding Ionos Expandgiven for 1vCPU which peaks at 4.95 pages/second.
• The offer LWS cPanel L falls to 1.75 pages/secondindicating the start of rate-limiting on our test server. This offer is announced with 12 vCPUThis suggests that it could offer superior performance in a different context. We assume that rate-limiting is in place, preventing our test from measuring the full potential of this hosting.

10 connections - ApacheBench - No cache

With 10 simultaneous connectionsthe gaps are still widening:

👉 The results observed are as follows:

• LRob Ultra retains the lead with 34.82 pages/secondfollowed by LRob Super with 31.64 pages/second.
• o2switch Cloud is now in third place with 19.78 pages/secondjust ahead LRob Pro à 18.07 pages/second.
• HaiSoft Pro WP follows with 13.58 pages/second.
• LRob Starter shows a decline in performance with 9.53 pages/secondfollowed by HaiSoft Essential à 7.02 pages/second.
• Hosters reaching their limit or applying a rate limiting appear next:
  • OVH PERSO
  • Ionos Expand
  • LWS cPanel L (rate limiting probable)

20 connections - ApacheBench - No cache

With 20 simultaneous connectionsperformance continues to stand out.

👉 The results observed are as follows:

• LRob Ultra retains first place with 48.19 pages/second.
• o2switch Cloud moves up to second place with 33.34 pages/secondexceeding LRob Super which seems to be reaching its ceiling at 31.78 pages/second.
• LRob Pro follows with 18.00 pages/second.
• The rest of the standings remain unchanged:
  • HaiSoft Pro WP with 13.58 pages/second
  • LRob Starter with 9.53 pages/second
  • HaiSoft Essential with 7.02 pages/second
  • OVH PERSO with 5.83 pages/second (probable rate limiting or offer ceiling)
  • Ionos Expand with 4.9 pages/second (probable rate limiting or offer ceiling)
  • LWS cPanel L with 0.95 pages/second (rate limiting probable)

40 connections - ApacheBench - No cache

With 40 simultaneous connectionsIn this respect, offers capable of exploiting a greater quantity of resources clearly stand out.

👉 The results observed are as follows:

• LRob Ultra reaches its peak with 48.19 pages/second served uncovered.
• o2switch Cloud follows with 33.34 pages/secondin front of LRob Super à 31.78 pages/second.
• LRob Pro maintains a solid performance with 18.00 pages/second.
• HaiSoft Pro WP poster 13.69 pages/second.
• The rest of the ranking remains stable with :
  • LRob Starter
  • HaiSoft Essential
  • OVH PERSO (probable rate limiting or offer ceiling)
  • Ionos Expand (probable rate limiting or offer ceiling)
  • LWS cPanel L (rate limiting probable)
  • Ionos Expand (test failed)

🚨 Rate limiting

• The offer LWS cPanel L seems to apply a rate limiting à 1 page/secondThis limits the server's ability to handle a high load.
• The offer Ionos Expand stopped responding to the test server, with 79 % of invalid responses (codes other than 200).
  • This indicates that the test has probably exceeded the limit authorized by the host.
  • As a result, the invalid value obtained was not included in the final graph.

ApacheBench load tests - With cache - Unexpected results

Caching a website can boost its performance tenfold. Load tests with cache are designed to measure the maximum capacity of hosting facilities under optimal conditions.

But not everything went according to plan for this test.

The results really went in all directions on this test, making any analysis fortuitous.

We note, however, that the record obtained is 503 pages/s on the LRob Pro offer for the first of these load tests.
HaiSoft reached a maximum of 176 pages/s and OVH 226 pages/s in the highest scores.

The conclusion is that we can't reliably measure peak performance with our methodology because of the rate-limiting features of most hosting providers. A cache is therefore essential for high-traffic sites.

We remind you that the type of cache differs for Ionos and OVH, which may influence the results of this test even more.

Here, however, are the results for transparency.

ApacheBench test - With cache -c 1-200 -n 1000

The initial test was designed for 1000 requests with 1 to 200 simultaneous connections. Problem: HaiSoft and LRob are the only two hosting providers to accept this type of load from the single IP of our test server. For all other hostings, this triggered error requests (different return code 200) at Ionos, OVH and o2switch, with a strong rate limiting at Ionos, the complete blocking of our server running the test on the o2switch offer, and a rate limiting to 1 request/s on the LWS offer, making the test far too long, so much so that we stopped it.

The raw values of this benchmark are made available for download to anyone wishing to re-analyze them.

ApacheBench test - With cache -c 1-100 -n 100

So we ran the test again with 1 to 100 simultaneous connections for 100 pages, which still triggered rate limiting. However, we let the test finish. This time, only our Ionos hosting showed "not 200" errors from 40 connections upwards, but others clearly seem to be applying rate-limiting.

The raw values of this benchmark are made available for download to anyone wishing to re-analyze them.

ApacheBench test - With cache -c 1-10 -n 10

We then waited a while and went back down to a total of 10 requests from 1 to 10 simultaneous connections. This doesn't really show the potential of the hostings at all, and still triggered the rate-limiting of Ionos and LWS. o2switch seems to show real values on this test, but we still have our doubts.

The raw values of this benchmark are made available for download to anyone wishing to re-analyze them.