TortoiseLabs introduces multi-class CPU SLA to vServers

[kaniini]

Hello,

We are pleased to announce a fully-functional multi-class CPU SLA for the vServers.  This allows for the nodes to be used for a class of computation known as "bulk computing" without disrupting other tenants on the node.  Examples of "bulk computing" would include things like crypto-currency mining, hash calculation/cracking, etc.  This is all possible because of our usage of the powerful Xen credit2 scheduler.

There are four classes of SLA: domain-0 (which is to ensure all domain-0 events always preempt VMs), guaranteed, standard, and bulk.  Guaranteed is a manually set SLA level which gives four times the priority for the same memory tier, and has to be requested by the customer for a VPS.  This allows the customer to ensure that they get the necessary CPU resources they require for a project, such as rendering or transcoding and is done on a case-by-case basis.

Right now our system automatically detects and assigns the SLA level (other than guaranteed) based on the previous day's CPU usage.

Here is what it looks like when we profile a vServer instance which has been assigned the 'bulk' CPU SLA.  In the console window, we have a console session open to a vServer instance which has been assigned the 'guaranteed' CPU SLA.

http://www.youtube.com/watch?v=jNFvIE6FXkw

As you can see, the amount of CPU time available to the 'bulk' CPU SLA vServer decreases as we spawn instances of cpuburn on the 'guaranteed' CPU SLA vServer.  Once we kill the cpuburn instances, the 'bulk' CPU SLA vServer gets their CPU access back.

 

[BlackoutIsHere]

This looks really cool! I might have to try it out some time. 

 

[drmike]

This is pretty darn neat!

Is this sort of "CPU abuse" a special uptick charge per month?  See it is a case-by-case basis in your post.

 

[kaniini]

This is pretty darn neat!

Is this sort of "CPU abuse" a special uptick charge per month?  See it is a case-by-case basis in your post.

No, we don't charge extra for these cases usually.  I mean, obviously the option is there, and there may be a situation in the future where we consider it, but it's mainly in place right now for the purpose of ensuring everyone's particular service needs are fulfilled.

After DigitalOcean wised up, we started getting Primecoin miners, so we needed to find a way to solve that problem which was relatively fair.  I believe we solved that problem with our use of multi-class CPU SLA.

 

[Mun]

So what is allowed for mining? I assume there is a way to auto schedule it to always be bulk.

Mun

 

[blergh]

Impressive setup indeed. What makes me happy is that there is "new" knowledgeable people around that are offering a product that there is some actual thought, skills and heart behind. While this is not a "world first" of any kind, innovation tends to bring positive things along, for everyone.

Good, sane pricing too.

 

[kaniini]

So what is allowed for mining? I assume there is a way to auto schedule it to always be bulk.

Mun

Right now, a script runs every day which determines whether or not a server should be switched to bulk.  We are still working out some issues we found with it, specifically in relation to the old weighting settings being stored (meaning a reboot would put you back in your old SLA class).

We are planning to resolve that fairly soon.

As for mining, it's allowed with the caveat that you may only get a few % of CPU at times if someone else, not classified as bulk, is doing an intensive task, such as 'make buildworld' or what have you.

 

[jarland]

Impressive setup indeed. What makes me happy is that there is "new" knowledgeable people around that are offering a product that there is some actual thought, skills and heart behind. While this is not a "world first" of any kind, innovation tends to bring positive things along, for everyone.

Good, sane pricing too.

Would agree. It is extremely nice to see people who enjoy what they do and work hard at it. Truly stands out in this little market.