Portal Home > Knowledgebase > Articles Database > High reliability config 4vCpu or 8vCpu ?
High reliability config 4vCpu or 8vCpu ?
| Posted by Docop, 02-04-2016, 06:52 PM |
|
Hi
I would like to know what would be the difference in performance for a VPS with this spec below, running Centos6.7. No price related, only performance.
A- 4 vCpu, 6gig Ram
B- 8 vCpu, 6gig Ram
C- 10 vCpu, 6gig Ram
I had tested a web server, as CentosWeb Panel, and on a small 1vCpu with 3gig ram, it run quite better than a 3vCpu with 1gig ram.
What setup will be able to handle a huge load and stay up and running ?
I do had try a load test over the web, and the 2vCpu with 1gig ram just crash directly and no ip access later on. And was running mod_sec and cfs firewall.
Thanks
|
| Posted by MattF, 02-04-2016, 09:00 PM |
|
The concept of virtual cpus means very little, as it can be different things in different contexts.
On say AWS, Azure, Google Compute when you choose an instance with a specific CPU profile you can almost guarantee a baseline of performance, this remains relatively static regardless of neighbors and time of day, hence these are favorable when scaling at the point of CPU saturation (due to predictability). Although you can find for example outliers (possible degraded or some resource attack vector of neighbors that does escape bounds), the hip scaling stuff (of which there dozens of providers) basically will decommission a poor node that doesn't meet your baseline, for example say you expect a M3.large to serve 300req/sec (or other app perf metric) any time of day then scale up +1, shutdown, scale down bad instance.. This excludes T-* range instances.
On your other end of the scale your Linode's, DigitalOceans, Vultrs, insert-reasonable-vps-provider. it's less well defined, in Linode's case they stick a certain number of guests on each node (by plan), and CPU is resourced limited by time slice / guests., your performance can vary wildly based on neighbors demands and time of day but at least the lower end should still be good for most non 24/7 CPU bound workloads (i.e. on a bad day it's still fine for most with a quality provider - I host websites with both Linode and DO for example). This is the v. The lower bound will be much much lower than above, but the average/upper 95% available would appear to represent better value for money, but when dealing with CPU contention and scaling you're almost always interested in the lower bound.
And then your openvz and low ends hosts, well it's a kids game here, you'll see benchmarks and speedtests on day 1 that completely contrast after node is loaded and people start using it (~2/3months) sometimes by 1000x factor, equally they have so little control over resources it is usually suspend on specific load metric TOS etc.. But you may find your 10vCpu, 6gb for $5 here
You should really determine if your load is CPU-bound, for most its not even though they initially believe it is. A high load metric (and even a deceptively low iowait) can also mean a I/O bottleneck in many scenarios.
Last edited by MattF; 02-04-2016 at 09:13 PM.
|
| Posted by net, 02-04-2016, 09:17 PM |
|
It depends on the provider's setup. No matter how many vcpu you have, if they are oversold, the performance will be poor.
|
| Posted by Savio13, 02-04-2016, 09:21 PM |
|
Unless you get dedicated cores, the performance will vary depending on the neighbors etc like others have mentioned
|
| Posted by Docop, 02-04-2016, 10:57 PM |
|
Hi So If I resume, could it be possible, by having an higher vCpu will give us more share on the actual server.. ? Like, a system running on 1vCpu might got less ressource or time, vs the same one running over a 10vCpu.
Same idea as If i got half of all the Vcpu for myself over a Xeon system, i should be quite ok, vs another one running only on 1..
If that make sense.
thanks again
|
| Posted by MattF, 02-04-2016, 11:21 PM |
|
No. Unless you're looking at cloud provider that baseline CPUs such as AWS/Azure/Rackspace/GoogleCompute/Oracle.
Not a perfect analogy but looks at it like this, with 4 vcpus I'll give you 4 slices of the cake, with 8 vcpus I'll give you 8 slices, what I don't tell you is how many slices are available. Most hosts cut the slices as thin as they can (i.e. slicing the pie into 300 slices). I'll take 0.5 of a slice any day if there are only 4 slices. The guy giving you 8 vcpus may potentially let you eat all those tiny almost invisible slices at once though, if you pray to scheduling gods, but in reality probably not as you're relying on at least 1/2 (or 1/3 in case of 24 core on node) of other people to stop eating at precisely the same time.
In other words you're usually looking at a marketing bluff for 2 or more core < $10/mo, granted even Linode tried to pull this off once before realizing it stank of deception, food for thought, don't be fooled by X vCpus
Last edited by MattF; 02-04-2016 at 11:31 PM.
|
| Posted by domainbop, 02-05-2016, 01:00 AM |
|
Another way of saying it is the processing power of a xx vCPU VPS can vary widely from provider to provider (and it can vary widely even on different nodes at the same provider) due to the differences between resource allocation at different providers and resource usage on different nodes so it is possible for a 4 vCPU VPS at one provider to have significantly more processing power than an 8 vCPU VPS at another provider. About 2 years ago I did a comparison of the UnixBench test results of several "2 core" VPS's I was using at the time and the test results ranged from 702 to 3946 (test results were here https://vpsboard.com/topic/4748-dear...-buy-your-vps/ ).
There is no 'one solution fits all providers' answer to that question because it does depend on each provider's setup, and can even vary from node to node at the same provider. If you're planning on doing something that requires "a huge load" you should first check the provider's policies on heavy CPU usage, and you can also run a benchmark (or search for UnixBench of ServerBear test results) to get a rough example of the power available (just bear in mind that test results even on the same machine may vary depending on what your neighbors are doing when the tests are being conducted.)
|
| Posted by PowerUpHosting-Udit, 02-05-2016, 02:04 AM |
|
It honestly depends on and the performance would vary from node to node and company to company. If you have a busy neighbour and if your provider is overselling the node, you can expect a nasty performance doesn't matter if you get 1 vCPU or 10 vCPU while if the provider is limiting the number of VPS per node, then you can expect a much higher performance.
Realistically what I would look for is, what kind of virtualization technology is used. From some of the analysis, we did a while back. Linux VPS performs marginally better over Hyper-V virtualization technology (P.S. I am not even considering OpenVZ as they oversold big time).
Another very important aspect that you are missing out is what kind of CPU is being used, faster processor will traditionally perform better. For example, avoid hosting companies that uses anything higher than 12 thread servers for VPS as they would usually oversell big time and that affects the network performance as well since majority of the providers are on 1Gig ports while they don't restrict and guarantee you a certain amount of bandwidth.
Summary:
- Go for KVM Virtualized VPS for Linux
- Look for Higher Speed CPU (3.3Ghz+ per CPU)
- Ask your provider how many VPS does he caps per Node
- Don't go for random unknown cheap hosting company
|
Add to Favourites 
Print this Article
Also Read
Shared Hosting
Shared Hosting
