Guaranteed cloud resources

Guaranteed resources

In theory ...

There are many providers of cloud infrastructure today but their services can differ substantially. We at BitSwarm know that very well as we have rented resources elsewhere prior to launching our own cloud and the resources we were promised to get sometimes differed considerably from what we paid for. With this experience and a thorough understanding of why and how this happens we are now offering guaranteed resources to our clients.

The guarantee means that the amount of resources you rent will always be available for your consumption but mostly you will be able to use more than you have reserved. The latter can be either when other clients have reserved resources but they do not use them all or when there are unassigned resources in the cloud – practically this is all the time as there are always more resources in the cloud as there are rented to accommodate future growth. The only thing that varies is the amount of non rented and thus available free resources.

The free (unused and unassigned) resources are distributed to clients by need and by a measure called weight. The weight is generally defined by how much resources a customer has already reserved – the more you have reserved the more free resources will be assigned to you when needed. Please mind though that the free resources are always available in a best-effort fashion.

The described model leads to the conclusion that the resources in the BitSwarm cloud are never oversubscribed.

In example ...

The concept of guaranteed resources is best described in a practical example. Suppose that we have acquired the IUs for 3000 MHz and we created a cloud machine and assigned it all the computing power available (namely 3 CPU giga-cycles or 3000 MHz). The cloud machine will now be able to consume 3 CPU giga-cycles at any time but in a moment it may require more. At that moment additional CPU cycles will be assigned to the machine if they are available in the cloud (best effort).

In case there is more than one contestant cloud machine for free cycles the distribution of these cycles will be weighted. If, for example, a cloud machine with 3 giga-cycles assigned and a cloud machine with 6 giga-cycles assigned would both need additional cycles, one third of the free cycles would be assigned to the former and two thirds to the latter. The formula is such that all the assigned cycles are summed and then each is divided by the sum to form the weight of the cloud machine. Example:

  • weight for cloud machine 1: 3/9 * free cycles = 1/3 * free cycles
  • weight for cloud machine 2: 6/9 * free cycles = 2/3 * free cycles

At some other time, when a cloud machine does not even need the CPU cycles it has assigned, they will be assigned to some other cloud machine. In this way not only a guaranteed minimum amount of CPU cycles (in our case 3 giga-cycles) will always be available to the cloud machine but additional cycles might be available as there are always some free cycles in the cloud – either because of under-subscription or by underuse of resources.

The same concept as described above for the CPU also applies to storage bandwidth, I/O operations and cloud internal network traffic.

In practice ...

Processing power

The processing power is available exactly as described in the above chapter:

  • CPU cycles rented in the form of IUs are guaranteed.
  • Additional cycles are available by weighted best effort.
  • All the additional cycles are free of charge

For the definition of the infrastructure units (IU) click here. More information about pricing can be found here and billing is described here.

Memory

The main memory of each cloud machine is always:

  • real computer RAM – The RAM of host machines is never oversubscribed so the main memory is never swapped outside the cloud machine itself.
  • available only as much as it is rented – there is no best effort over-usage of unused RAM in the cloud.

For the definition of the infrastructure units (IU) click here. More information about pricing can be found here and billing is described here.

Storage

  • All storage is persistent – This applies to cloud machines, images and backups - everything labeled storage in the BitSwarm cloud means it is persistent.
  • Storage space is only available as much as it is rented – no over-usage is available
  • Storage bandwidth (the amount of data that can be read and written) is guaranteed to be 400 KB/s per IU. More bandwidth can be gained best effort as described in the above definition of guaranteed resources or buy renting additional storage units – each storage unit includes an additional 100KB/s of guaranteed storage bandwidth.
  • 25000 I/O operations per hour per IU are guaranteed. More can be obtained in a best effort model as described above. Additionally any rented (and assigned) storage unit adds 10000 guaranteed I/O operations per hour.

For the definition of the storage units (SU) click here. More information about storaga pricing can be found here and billing is described here.

Internal network traffic

  • 400 KB/s per IU of network bandwidth for data transfer between cloud machines is guaranteed.
  • The bandwidth is aggregated for all cloud machines in the BitSwarm cloud and is not specific for each cloud machine.
  • The total data transferred is not limited.

Network traffic is described here. More information about pricing of the data transfer can be found here and billing is described here.

External network traffic

  • All the data transfer rented in advance will be guaranteed. The network bandwidth guarantee will be the rented amount distributed evenly throughout the month and multiplied by 2.5.
    For example: Renting 3TB of data transfer would mean a guaranteed bandwidth of 22.83 Mb/s. [amount of data transfer reserved / number of seconds in a month * 2.5 * number of bits in a byte] = [3000000 MB / (3600 s * 730 h) * 2.5 * 8]
    There is a possibility to use more bandwidth at any time but the availability is best effort.
  • The rented data transfer applies to both inbound and outbound traffic. In the above case the guaranteed data transfer would be 3TB inbound and 3TB outbound and the bandwidth would be 22.83 in and 22.83 out. Only the outbound traffic is billed though. See the chapter billing and payment for more information.
  • The data transfer needs not to be rented in advance. In this case the bandwidth is provided best effort. This option is suitable for minor use as the best effort will always suffice.
  • When opting for the second model – reserved bandwidth and bursting (as described here), the guarantee is for the reservation but also for the possibility of bursts. Contact us for more information.
  • The data transfer is always aggregated for all cloud machines and is applicable per customer not per cloud machine.

Network traffic is described here. More information about pricing of the data transfer can be found here and billing is described here.


↑ Top