11My preference has been with the virtual private server in the dedicated vs. VPS debate, however sometimes I do wish I am on a dedicated server, where the resources are shared with no one but myself alone. Here is the evidence I have taken from one of my VPS a few weeks ago.
It is generated by Cacti running on another server, sampling SNMP data every 5 minutes and ploting various graphs. This graph shows a big “spike” of server load up to 7 from 10pm to roughly 11:15pm. Then the load dropped back to normal again because of a reboot.
You must have been wondering what was the VPS doing in the Friday evening. Nope. It was not drinking in the pubs. Nor did it go clubbing. It was humbly sitting there inside a dedicated server, serving requests at average less than 1 request per second. No strange traffic. No DDoS attack as far as I am aware of. That VPS was pumping out at only 15kbps. Running top showed that there was not even a process that is hogging the CPU.
On another VPS where I have periodic load spike. Check out this graph:
Basically, after afternoon from 3pm to 5pm Australian Eastern Standard Time there will be a load spike, and I am very sure that I do not have a cron job running then. It looks like it only went up to 0.3, but only because it is an idle server which I am getting rid of soon. When I was actually using it two months ago, the load spike has been much worse. And I learnt I should not log into that server to do any work in the afternoon because even doing basic command line stuff can be unbearable.
Bad neighbours doing weird things at weird time — they don’t just exist in dodgy suburbs, but they are there sharing the same physical server as your VPS as well!
I myself am guilty sometimes. Yes, a few Linux users I know love Gentoo Linux, but hate to be with other Gentoo users as VPS neighbours. I’ve just done a emerge --sync on one of my VPS, and CPU was 99% in waiting state most of the time due to the amount of IO required.
D’oh.
I thought vps was more protected or is this mostly from the disk io being used by other users, or by the cpu as well?
I think it could be both, as some VPS providers do over-subscribe their physical servers. Also it can be such a pain when everyone’s cron job doing slocate indexing at the same time — mostly due to people installing the same Linux distribution and using the same timezone.
IO is certainly going to be the biggest bottle neck for VPS. Intel already has 4-core CPUs, and from the look of it, scaling cores in CPUs is going to be the future. Maybe by this time next year all the CPUs you buy are going to be multi-cores. However scaling hard disks is not going to be that trivial. There will be bigger and faster disks, but doing 4 disks or 8 disks RAID to match cores in the CPU is not going to be cheap.
As of IO scheduling and prioritisation in virtualisation software — I don’t think they have reached perfection yet.
I’m one of those Gentoo people that do things at what would be an inappropriate time :P
On my GPL Host I get stuck with I/O wait a lot of the time but on my VPSLand one I’ve never had an issue.
Both running Gentoo of course.
For example the other day I was just trying to view a couple of logs “less -S /var/log/em<tab>” the I/O wait locked me out and i couldn’t even change windows in a screen session, crazy…
Its unfortunate that I need to be on a AU ip for this project. Oh well :)
But at the same time I wouldn’t want to be on dedicated… VPS is so much better from the viewpoint of what I have to take care of. As long as the O/S & apps are good, I don’t have to worry about the hardware, that’s someone elses job. Considering the logistics involved if something failed, no thanks.
Be nice if everyone could afford proper SAN’s to back their CPU power, but then I guess their price wouldn’t be as low would it…
Yup, I have got some pretty bad IO wait on GPLHost as well. Their dom0 graphs are not very helpful here as you can’t see which domU is chewing up all the IO :(
Hi!
Just to tell that this is about to change as we are currently addressing the problem with a daemon that fetches the I/O stats for each domU directly in our control panel. We will put limits to our customers so VPS wont swap too much.
Thomas
Ooooh bad news this weekend.
Someone on my node is maxing out the server and the website that brings in leads is not working… Its still up but since it can’t respond due to no available resources it doesn’t work. If we don’t get leads, don’t make money, don’t make money, don’t have a business/job.
Going to have to keep this site outsourced I think, which defeats the whole purpose of me being here… Yeah I’m doing other things.
If only data connections were affordable to the average punter, poo….
Wonder if I should bother with another VPS provider or just look into dedicated….
Well there is definitely one thing GPL Host have going for them and that is their customer service!! I sent of an email just before I posted here and waited for something to happen and one of the guys came on IM.
So it’s all been sorted out. The resource hog has been slain and the site is back up :)
Well. Maybe I wasn’t so lucky then. I think I am on the same node as you (node6101).
I woke up this morning at 7am, and found none of my websites on that VPS is working. Found out about the bad VPS on that server via Xen network statistics provided. Went online and no one was there on IM, so I filled an email at 8:30am about the issue.
Looks like it wasn’t resolved until 1pm AEST.
Worse, I went out to a have a barbecue at 10am, and did not come back home until JUST THEN! All my PHP FastCGI processes are in zombie state so even though the VPS is responding, my visitors are still getting gateway error from Nginx! It would not be an issue if they just reboot the box, but then I guess it is beyond their control.
So 19 hours outage for an otherwise quiet and relaxed Saturday (great weather in Sydney today, and it was a great barbie :)
Yeap same node :)
Indeed the weather has been great this weekend in Sydney.
D siad he was going to look into preventing this sort of thing in the future.
Later in the afternoon for some reason (unrelated to earlier incident) the path from my GPL Host VPS & work (bigpond) to my other vps in Atlanta got blocked. So the email form on the site at GPL Host wasn’t working (don’t have sendmail installed just ssmtp forwarding to postfix in Atlanta). Such a pain in ass. I’ll be looking into putting the form data straight into a local mysql table today….
So much work to be done for everything to be working properly and it all needs to be done yesterday….
Oh Well, means I have job I guess :P
There is very new I/O priority support in the latest OpenVZ builds. It’s as simple as running vzctl with–ioprio option on a VPS :) This functionality will make it to Virtuozzo in 4.0. So far it looks promising.
Thanks Matt. We actually have been testing the latest vzctl at work and spot the new IO options. I will surely be looking out for VPS companies providing the latest OpenVZ and the up-coming Virtuozzo 4.