"Safe" Overselling?


#21

Hi so…

Can anyone tell me if this is normal.
It’s my first time running a Node and solusvm on it.
Server has 64gb ram and 2x Xeon E5640 6c each and has 6 OpenVZ containers.

I also noticed that 1 client is using all that

but this client only has 4gb of ram and 2 cores alotted.
Is it normal for complete node to see that spike or not?


#22

Client is probably reading/writing a ton to disk.

Try limiting their container’s IOPS or disk access/write speeds to correct.


#23

Burstcoin.


#24

I see some spikes on disk, I’ll limit him.
Thanks.


#25

I/O is a shared resource, so someone heavily reading from or writing to the disk can definitely affect the performance of the entire server and cause the load average to spike (particularly if the load average spike is due to iowait). This is especially the case with containerization technologies like OpenVZ and LXC. Limits should help :slight_smile:


#26

Solid and reliable post.
Highly recommended.


#27

Thanks @Daniel, the problem is still present for me. I changed the IO priority from 4 to 7 in solusvm but tha didn’t help. It’s openvz with master on the same node. Do you know if I can put that vps on a hard limit somehow? or Will limiting it’s connection speed help?


#28

OpenVZ shares a common load value, so if one VPS makes it load go high then will be reflected node wide.

It’s hard to put hard limits as it more works on the basis of % of available spare resources, so if your other VM’s are doing not much then the load average on the host will basically be down to the single VM. Only things like CPU and RAM can be truly hard limited with VZ

Is this a VM your running? If so is there a high CPU usage within the VM?


#29

There are not many nodes in the node but I can see the problem and I confirm that it’s the disk IO and not the ram.

I might move the client to a dedi or a kvm when I get a new node but for now I think I just have to ask them reboot their vps once or twice a day


#30

Not enough RAM leads to excessive swapping and when that starts, there is no return.


#31

but should swap with OVZ not be vswap aka ram that’s throttled?

you could try and up the RAM for that client to see if it helps. maybe try to clarify on his use case, and point out that this affects the whole node and he needs to use another product for that. eventually offer to refund him to get rid of him soon.

however, I’d also strongly advise against OVZ anyways.


#32

He has 4gb of ram, 8gb burstable and 4gb swap none of which is being utilized over 50%.
Ram usage spikes once in a while but then it settles back to around 2-2.5gb usage.


#33

Yes i did that. Offered him a full refund and a discount on a dedi. I’ll have to contact the DC to add another server in their racks for me which can take upto 5 days. So until then I’ll just have to live with the load and hope rest of the clients don’t get affected enough to notice it.

And I agree on staying away from openvz, but with my budget since I just started 20 or so days ago I’m using same node for master and virtualization which limits me to openvz. My next node will be KVM and probably SSD instead of SAS


#34

Which software? This should never be true.


#35

OVZ must die!


#36

Do you know what they’re running on the VPS?

Load average is the number of processes that are actively running, plus the number of processes that could run but are blocked for some reason - usually waiting for a shared resource like IO or CPU. Load average is given as three numbers - the average over the past one, five and fifteen minutes. A load average of 1.00 generally means that one core has been fully utilized (100% CPU usage) for the entire past one, five or fifteen minutes. Because of that, you don’t want the load average to exceed the number of cores on your server. The load could also be caused by processes waiting for the disk - you can check that in “top” by looking at the iowait percentage (listed as “wa”).

A load average of 113 means that 113 processes are either running or waiting to run. That’s ridiculously high. It’s possible they’re running a lot of stuff on the VPS, all hammering the disk, so there’s a giant queue of processes waiting for the disk or CPU.


#37

Both solus and virtualizor have that limits. I can only run openvz if master is on the same node


#38

Hey Daniel, Yes I was aware of that.
They are trying to use this vps as some sort of cdn so that writes and reads a lot of data at the same time. Most of it was video content. I talked to them and they showed that 183 processes were running at that moment. I provisioned them a dedi server now so waiting for them to leave the node and also provided him a full refund so that there are no hard feelings. Their other VPSes are fine.

That’s the top output

top - 13:09:30 up 16 days, 22:39,  2 users,  load average: 19.00, 17.81, 17.83
Tasks: 1004 total,  17 running, 987 sleeping,   0 stopped,   0 zombie
Cpu(s):  9.4%us,  0.6%sy,  0.0%ni, 89.9%id,  0.0%wa,  0.0%hi,  0.1%si,  0.0%st
Mem:  65804760k total, 41929384k used, 23875376k free,  1576448k buffers
Swap: 69206012k total,     1508k used, 69204504k free, 31790380k cached

#39

What you could do, with Virtualizor at least, not sure about Solus.

Install a stand alone KVM server, create a KVM VPS, install the master in that KVM, You could then add the Node Virtualizor as a slave :slight_smile:

When needed, you can move the Master KVM wherever you like, and update the IP on the slaves.


#40

Or just install it on a 3 EUR Hetzner VPS.