I have experienced too many times that CPU-steal went high from very low, causing my scripts to run slow and every time the Linode-staff are like "uhm, nothing to see here - wanna migrate?".

Migration takes around 3 hours, without knowning when it will happen and it really only delays the problem a couple of weeks.

Today it happend again (hence this post). CPU-steal went from around nothing to 30-35% in a minute and it's still there after 8 hours. My scripts are running approx. at half speed. Linode-staff are like "We've checked it, everything is OK .. wanna migrate?".

They always respond very quickly, which is great. They also made some adjustments which they tell me I will benefit from in about 24 hours. This is also great - but the lack of tools to see what's really the problem is an issue for me.

Therefore; you need better tools guys!

CPU steal isn't really a problem. You're in a shared environment that allows you to use more than your equal share of your CPU. You will always be able to get your equal share, if your app can't run on it's equal share then you need a bigger node.

_________________
Paid support
How to ask for help
1. Give details of your problem
2. Post any errors
3. Post relevant logs.
4. Don't hide details i.e. your domain, it just makes things harder
5. Be polite or you'll be eaten by a grue

You buy 1 CPU share and Linode give you 8 CPU shares by giving you access to what other customers have paid for. You feel hard done by when these other customers use their CPU share.

Unpredictable amounts of CPU time and IO bandwidth is a feature of all virtual servers, but it's not something that most people notice. If you really can't live with this you need a dedicated server not a virtual server. Most applications run fine on virtual servers though, even sensitive stuff like asterisk.

It's not the CPU-steal thats the problem. It's that the people at Linode cannot see a sudden change in user behaviour. My node is fairly large and quite oversized to make sure my site is running smoothly, although it isn't at the moment.
To make this jump in CPU-steal I suspect that one other node must be running all 8 cores at 100% and I would like a change like this to be visible to my hosting provider.

I would have preferred a response like this "Yeah, some guy is maxing out all cores, but he paid for them. Live with it". Then I would know what to expect and scale my site to whats needed.

You're right btw, a dedicated server might be the right solution for me.

It will be visible to them, you said they say "We've checked it, everything is OK .. wanna migrate?". If someone is using all 8 virtual cores and maxing them out then everything is OK so what they're saying is fine.

_________________
Paid support
How to ask for help
1. Give details of your problem
2. Post any errors
3. Post relevant logs.
4. Don't hide details i.e. your domain, it just makes things harder
5. Be polite or you'll be eaten by a grue

For the record; the steal-level was (and still is) abnormal, the responsible Linode is identified and they are trying to work out a solution. I guess the level of service depends on which supporter you get hold of.

I have the same problem for about 2-3 month now. Some time in Nov steal level went up to 10-30% and stayed there, sometimes peaking at around 1oo%.
I talked to the support guys, they were very quick to respond but when probed about what's going on on the host maschine, all they could tell me is "no idea what all the linodes are up to and who's using up all the resources. want to migrate?".

looper - what did you end up doing? Stay on the host? Migrate elsewhere?

I wonder if the move to 8 core virtual machines was a good idea for Linode, as sometimes I'm getting awful performance myself.

Server was great until this point (2pm).
Migrated it myself to another host server about 10 days ago, things are now better, but not was great on the 1st server.
And now yesterday and today - things are worse again.
Data from munin on the server itself seems to confirm it as well, process time doubled until I migrated it myself, and now things are looking kinda crappy again.

I understand that these are virtual machines and performance isn't guaranteed, but it's slightly frustrating getting such a drastic performance drop on a server which performed so well in the past - especially when you use a small amount of resources which feels proportional to what your fair share should be - but you don't get it.

My personal solution is to migrate the data to a new host server myself and swap the IP addresses once I'm done. It's not something I like to do, but it's faster than a Linode migration.