New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
that depend what processor you use, and what hdd do you use. the bottleneck usually in hdd or processor not ram
Just updated the question
hdd? use raid 1,0,10? SSD,SAS,SATA?
It depends. We have undersold (biz plans), approximately balanced (the GB sold are close to the physical GB of ram, it varies, some people insist for a VPS and if the server is doing fine with less than 30 % cpu load and less than 40% disk used, we create a new slot as well for some upgrades), and have clearly OVerZold ones, there the percentage goes over 150% at any time, closing to 200% on servers with less memory where the CPU has no issues.
Since we allow heavy CPU game servers and transcoders, etc, we cannot oversell much since the CPU will fail way before RAM is exhausted it is the case on the OVerZold plans, some have 20%+ free ram (which is good fo caching, dont say is bad) but the CPU hardly has 4-5 threads left (out of 24) at peak times leaving little room for abusers. People which only allow usage of one of 4 cores for example and limit gameservers and whatnot can oversell much more, I think 400-500% is not uncommon.
That depends entirely on the provider. There are some providers who sell 150GB-200GB of RAM or more. on a 32GB node
it all depends on your CPU and your hard drives you can have best CPU in the world with 32GB ram but with standard disks with no raid cache you IO be really high and cause a bottleneck.
for unethical providers such as U**** , S******** They must go over 200% at least to make profit. why? because the performance is shlt like 200% or more
I would suggest between 120 and 200 is probably closer to normal and perfectly doable.
As many as the IP addresses you have
Squeeze them all in, huh? :P
I would sell about 25, 25GB sold, then that leaves 2GB For the OS and 5GB for burst/upgrades etc
128
As someone else mentioned, you will probably run into I/O bottlenecks before anything else. It depends a lot what you are doing and how your I/O is set up
If you have SSD's or RAID10 you could easily get 40 or more. You will most likely still have plenty of actual memory left after that and will not be swapping. That is because even though you allocated 1GB it is absolutely certain most of not all will not be using all of it which leaves it available for use.
If you are not doing anything special with I/O beyond software RAID1 I would say 40 would probably be pushing it. I would suggest you consider only about 20-30 users and downgrade to 16GB if it saves you some money.
Your question is too vague. Who is going to be using the OVZ (is it a private thing to separate off your own software, or are you selling it to clients directly as IaaS)? What software is going to be running in each container? What's the expected network usage / I/O usage / CPU usage per container for the software that you want to separate into VM's? Or if you are selling it, what kinds of requirements do your customers have?
The best way would be to test various setups with your software and see how it performs and how it loads the host node.
Wow.. who in there right mind would cram that many VPS's into 1 node? Oh hold up..
It can be done of course, but that's to the extreme and you would indeed find your CPU/Disk's being the bottleneck unless you went with a high end storage.
I wouldn't even put 200 VPS's on a 64GB system
@Reece to answer your question "who in there right mind would cram that many VPS's into 1 node?" A lot more than would admit to it.
I have about 250 on a node with an X3210, 4GB Ram and 1 disk, and they are in use so with 32GB and an E3 I would say 120 - 200 is being modest, I strongly suspect that with an SSD pretending to be ram for swap maybe 5 - 600 would be the point at which 'most' customers would complain.
just my 2c
So imagine if I had an SSD cached raid array and an E3 behind it..
Now do you do a survey before handing selling VPS (if you do? ) I just asked for your view, how many VPS can you imagine selling off a 32GB server.
Imagine?
I would love to see someone hit 1000
Please don't post our dirty little secrets of OpenVZ . Let the "OpenVZ sucks because I read on some blog site from some guy who hosts Xen servers that..." crowd continue to believe their hosting myths.
Haha but what's the size of those VM's? There not 1GB! Probably 128MB or less.
@Reece you are correct however higher ram scales in the hosts favour as generally less is used with my VM's it is hard not to use all the Ram. with 1GB the chances are at least 60% of your customers are using less than 512MB on avg.
if you have 32GB (8x more than my node) and raid 10 with probably 8 x the IOPS that I do then why not 8x the VM's ... oh wait 8 x 128 is 1024.... so I stand by my point 200 is fine with OpenVZ on a 32GB node
It is no big secret really, but you always see the same coments when this same question gets asked, CPU bottle neck (rare and can be dealt with), Disk I/O (again if managed and set up correctly it is not going to be an issue especially on OpenVZ without LVM and virtual block deviced like with Xen and KVM), and finally Ram... OpenVZ + Ram = you have to be doing something very wrong to run out
Sorry @sman I think the secret is out. I hate to use this example and perhaps it is not in good taste but CVPS said they had about 50 nodes at one point and when the DB was leaked people said 10k+ active customers and this was pre E5 iirc so do the math, 1000/50 I am not saying this is what they do as I could not possibly know, I am simply saying it is viable.
I would prefer people believe the myths. In lowend land it's not much of a secret because OpenVZ is the defacto standard. Not so much in other things.
If someone's margins are so low that the small difference in price between a server with 32GB and one with 16GB is the difference between them turning a profit or a loss I would suggest that they rewrite their business plan.
from an end user's perspective, who in their right mind would sign up for a plan knowing that the 32GB node is that oversold and they'll be treated to mediocre performance, frequent disk I/O slowdowns and other performance problems? None of the >1GB openvz VPS's I currently have are hosted on 32GB E3's because while 150GB-200GB on a 32GB box might be doable for the host, more times than not it results in a crappy experience for the end user.
And that right there is the key
If someone's margins are so low that the small difference in price between a server with 32GB and one with 16GB is the difference between them turning a profit or a loss I would suggest that they rewrite their business plan.
Sometimes the difference is significant. That is changing but memory traditionally was one of the upsell things a lot of Dedi providers would try whack people on.
Of course if you have your own servers it's not so much of a factor. Also 32GB is now getting more standard. It depends on the guys situation which we don't have all the details on.