When starting your servers with 100% CPU usage, are there any problems or is it just good CPU usage?

Question

Does your server support 100% CPU usage, any problems, or is it just good CPU usage?

About 100% of the "open hours" / 10 hours for one salary are constantly working on my servers of 8 physical cores.

The program has an architecture for working on 8 threads and makes full use of them. Performance is good, but infrastructure partners worry about "maximum servers"

I think this is just a good use of the resources available. What is the point of having many cores if they are not all fully utilized.

Answer 1

All answers to date relate to the software of this problem. I think the OP may be interested in knowing if such processor usage will affect the reliability of the server hardware. Or at least I feel that I can give a fairly accurate point of view on this side of things.

The reliability of many server components is highly temperature dependent. In this regard, a server with a low load is likely to work colder and has a lower probability of failure. But as soon as the processor reaches the maximum allowable temperature of the cooling system (on my PC - 60ºC), the reliability will not be much lower at higher loads, because the cooling system will adjust the temperature to avoid overheating of more than 60ºC. It is true that the cooling system will make the fans move faster, which will increase the likelihood of fan failure. There is another problem with such a high load, if heat is not removed from the chassis in an efficient manner, the entire server will have a higher temperature, affecting other components. The components that will experience this to a greater extent are electrolytic capacitors, but other components will also be affected.

Thus, in general, yes, the realism of any machine is highly proportional to the excess or safe ratio of this machine. There are many formal methods for calculating the reliability of electronic systems, one of the most famous is the reliability manual MIL-HDBK-217F, but there are many others (Telcordia, etc.). Reliability is a topic with great complexity, depth and the slightest magic.

I hope I was able to talk a little about this topic.

Answer 2

Almost without exception, it causes problems or may cause problems in the future (as demand grows). 100% CPU usage on the web service server is not very good.

If your processor is working at 100%, this means that every time the server receives a new request, there is a 100% chance that the work will have to wait a while before the server starts working on it.

A typical characteristic of a sweet spot is about 70%. Does this sound bad? If so, remember that 70% of the usage does not mean that 30% of the CPU is wasted. Instead, this means that 70% of the processor capacity was used during the sampling period. For CPU readings, the sampling period is approximately 2 seconds. During these 2 seconds, the breakdown of these 70% is uneven; in other words, it could be something like 100% in 1 second and 40% in 1 second. 100% utlyization in short bursts, which is good, because we know that if part of the work is delayed, it will only be for a short period (the one that does not make the person who is waiting for it is upset).

Answer 3

Depends on the load. If the load is some data that does not grow - you are fine, you just do everything as fast as you can, and that's good.

But if the load comes from outside - for example, when the server is similar to StackOverflow requests and service client requests - this is the same, because you need to process requests in real time. Think about what happens if the load is increased after the sale announcement. Most likely, some requests will be deleted, while others will be processed slowly, and this will at best lead to the fact that customers will work poorly.

Answer 4

What do the servers do?

• If they solve math problems, than you want to get every bit of processing.
• If they serve actual users, you may want to have a small room for random peaks so that it does not break when your user base grows by 10%.

Also: What does the CPU do? Is it plausible that they really have to calculate so much? How much I / O should be involved?

The rule of thumb that I saw is that the maximum load is 80% average - less if you expect high usage peaks

Answer 5

Firstly, low CPU consumption reduces power consumption. Processors automatically reduce their frequency when they are not under heavy loads.

In addition, the presence of relatively low CPU utilization means that latency will be low, since each task will be interrupted as little as possible (on the contrary, high CPU utilization causes many context switches).

And if you need to run several applications, such as a database server, a web server and a mail server, then low CPU usage makes it possible to work for all tasks.

Surprisingly, there are huge differences in HTTP servers regarding their architecture and implementation, see http://goo.gl/Ha2rt

The right choice can help save money, avoid premature equipment failures and, please, users with less delay.