Advertisement IC Markets

Noisy neighbors on a shared server

Optimize MetaTrader 4, MetaTrader 5, cTrader, and TradingView for speed. Discuss Level II Market Depth (DOM), custom hotkeys, and volume indicators.
Post Reply
Fairman
Posts: 7172
Joined: Tue Jul 21, 2026 7:11 am
Location: Abuja

Noisy neighbors on a shared server

Post by Fairman »

Noisy neighbors on a shared server

A cheap VPS shares physical hardware with other customers. If one neighbor runs something heavy, the others can feel it. The effect is called the noisy neighbor problem.

The symptom is variable performance. For an hour, everything runs smoothly. Then the platform lags, orders take longer, and CPU metrics show spikes that no process on the machine explains.

Providers differ in how they allocate resources. Some guarantee a share of CPU and memory. Others oversubscribe, selling more capacity than exists, on the assumption that not everyone uses it at once.

A metric called CPU steal time reveals the problem on Linux, and some Windows monitoring tools show something similar. It measures how much time the virtual machine wanted to run but had to wait.

For latency-sensitive trading, reliability matters more than peak speed. A dedicated or reserved-resource plan costs more but delivers steadier timing.

Testing helps: record the response time of a simple benchmark every minute for a day. A stable line is good. A line with random spikes is a warning.

Over a day, what does the VPS performance log look like, and are there spikes with no explanation?
It’s Fairman :geek:
Recommended broker for automated trading & scalping IC Markets
FXS HongKong Scalper
Posts: 618
Joined: Sat Oct 24, 2026 10:15 pm

Re: Noisy neighbors on a shared server

Post by FXS HongKong Scalper »

Fairman wrote: Testing helps: record the response time of a simple benchmark every minute for a day.
Running a benchmark at different times of day can reveal when neighbors are busy. If performance drops regularly at specific hours, moving to a dedicated or better-provisioned server may be worth the cost.
FXS Official Team | Hong Kong Session Desk
Fairman
Posts: 7172
Joined: Tue Jul 21, 2026 7:11 am
Location: Abuja

Re: Noisy neighbors on a shared server

Post by Fairman »

FXS HongKong Scalper wrote: Thu Nov 19, 2026 11:12 am
Fairman wrote: Testing helps: record the response time of a simple benchmark every minute for a day.
Running a benchmark at different times of day can reveal when neighbors are busy. If performance drops regularly at specific hours, moving to a dedicated or better-provisioned server may be worth the cost.
That is fair, and the time-of-day angle is the part most people skip. A single benchmark run tells you very little, because a noisy neighbour is usually not there when you happen to look. What I would add is to keep the log from the same hours across several days, so a spike at the same time each evening stands out from a one-off blip. If the slowdowns line up with the hours you actually trade, that is a real cost. If they land at 03:00 when you are flat, it may not matter at all.

I would also say the cost question is about what a bad minute costs you, not about the monthly price. A dedicated plan that is a few dollars dearer is cheap if one delayed modify per month would otherwise cost a full stop. If you never hold anything open through the lag, a cheaper plan with a good log may be perfectly fine. The log is what lets you decide with numbers instead of a feeling.
It’s Fairman :geek:
Amanhasnoname
Posts: 2245
Joined: Fri Oct 30, 2026 9:45 am

Re: Noisy neighbors on a shared server

Post by Amanhasnoname »

Steal time is the right metric to watch, and it helps to know what normal looks like. A few percent now and then is common on shared hardware. If it regularly climbs into double digits, the host is oversubscribed and your platform is being made to wait, no matter how light your own load is. On Windows, where steal time isn't exposed directly, a simple timer loop that measures how late it wakes up each second gives a decent proxy.

Before blaming neighbours, rule out your own machine. Windows updates, antivirus scans, a browser left open with heavy pages, or several terminals each running indicators on many charts can all produce the same spiky pattern. Task Manager or Resource Monitor during a spike will usually tell you whether something local is eating CPU or disk.

It's also worth separating CPU delay from network delay. Log a ping to your broker's trade server alongside the benchmark. If the ping stays flat while the benchmark spikes, it's the host. If both jump together, it may be the provider's network or routing, and a different data centre location could matter more than a bigger plan.
Post Reply