Maximum Server Load After Which You Should Upgrade

[firstimpression]

I wonder if anyone can give me an opinion regarding the following issue. 

 

What is the horizon of CPU % utilization and system load, after which the load begins to negatively affect delivery latency?

Let's say, we talk about Linux system with 4GB RAM and 4 CPUs, using memcached, without GUI (only delivery)

[waterwheel]

Bit of an old topic, but there's no specific number for this.  Instead, try using http://www.webpagetest.org/ . to check the speed of your websites and determine if there's a speed problem and if so, where. 

 

For example, one of the things I do us use a professional DNS system that has DNS servers located all over the place.  DNS lookups are then done close to the visitor, speeding things up quite a bit.

 

I also am colocated in a facility that has peer sharing with most of the main ISP providers in my country.  So when visitors access our sites, there's not many hops to get here. 

 

Both of those things can make a visibile difference without upgrading hardware.

 

More ram is always an easy fix too.  4g seems low, 16 gig would be better.  And failing that, a simple bump to an SSD hard drive can work wonders as well. 

 

Most of the other hardware stuff will likely have a small impact compared to all of the above.

[andrewatfornax]

Totally agree with @waterwheel - asking "when will latency be affected" depends on the question "what do you want the latency to be"?

And with what kind of tolerance, and distribution of responses?

If that matters to you, then you need to be monitoring the performance of your site's delivery, and acting accordingly.

Admittedly, that's not a trivial task, but is, unfortunately, well beyond the scope of what the RA team can advise on!