Virtual server processor load due to utilization

When there is not enough CPU power available, because of to the modest vcpu level of the virtual server, the load on the application server starts to show up in the CPU utilization, which is visualized here in the graph as it is shown in the Hetzner web interface. It shows that instantaneous CPU utilizations have been considerably higher than they usually are.

The same level of load can also be characterised in e.g. the Datadog monitoring service, which in this case shows how increasing the number of visitors per minute by a thousand, and again by a thousand, etc., has resulted in an ever-increasing load.

Excessive queuing of visitors causes delays in processing

test page: writing with a couple of dozen paragraphs of text and a few images

page loads: 8000 per minute

vcpu: 4

maxthreads: 3 - 4

survivability: when many visitors are not prepared to be handled very much at once, visitors have to wait longer to be processed, but even setting maxthreads higher by just one can be enough to stabilise response times

Front pages of solutions often cause only minimal load

test page: front page of a solution, about forty writings in a few collections of writings, shown using front type of plain structure

sivulataukset: 10000 per minuutti

vcpu: 4

survivability: uncached page loads with a smooth response time of about 110 ms, with max. visitors. 200 per second

Simultaneous processing of image uploads causes wavering in response times

test page: front page of a solution, about forty writings in a few collections of writings, shown using front type of plain structure

concurrency: throughout the test, the same virtual server receives and scales one image at a time to different sizes having 1920x1080 pixels

sivulataukset: 10000 per minute

vcpu: 4

survivability: slight wavering in response times, but no more than about 40 seconds, when maximum number of visitors is 200 per second

Thousands of writing collections can be loaded with lesser server resources

test page: all the about 30 writings from a writing collection loaded at once

page downloads: 4000 per minute

vcpu: 4

survivability: response times remain reasonably low, but there is constant chatter

A few tens of thousands of writings can be loaded per minute by increasing the processing power

testisivu: a writing having twenty paragraphs and few images

page loads: 35000 per minute

vcpu: 8

maxthreads: 140

survivability: steadily increasing the number of visitors increases the response times quite correlatively for a writing that would be loaded seperately in about 80 ms, and the response times do not stabilize, but 35000 page loads per minute with a good average loading speed of 250 ms is not a bad test result at all

Nearly ten thousand writing collections can be loaded by increasing the processing power

test page: a writing collection having 30 writings loaded at a time

page loads: 4000 - 9000 per minute

vcpu: 8

survivability: response times improve by almost 100 milliseconds compared to vcpu 4 and remain more or less stable, staying at that level up to 7000 page loads per minute, but 8000 starts to become more difficult for the server and 9000 was then more or less impossible to test without timeouts growing very high

Steady increase in visitor numbers makes it difficult to stabilise response times

test page: a writing collection having 10 writings loaded at a time

page loads: 10000 per minute

vcpu: 8

survivability: doubling the number of vcpus from 4 allows to pass the test that last for a minute, but a steady increase in the number of visitors increases the response times quite correlatively

Timeout errors may caused by configurating server imprudently

test page: a writing collection having 10 writings loaded at a time

vcpu: 4

survivability: when a Tomcat server is configured in an imprudent manner, it can become badly overloaded, with visitors being forced to queue up and some not getting fully processed