Multi-pipe bandwidth control in hosted systems

A method comprising:

• receiving a write request from a guest process running on a host device, wherein the write request comprises target data;
• determining whether a token pool of the host device has sufficient available-tokens to write the target data;
• when there are not sufficient available-tokens in the token pool of the host device to write the target data, postponing writing the target data until there are enough available-tokens in the token pool of the host device to write the target data;
• when there are sufficient available-tokens in the token pool of the host device to write the target data:
• breaking the target data into multiple write blocks;
• writing each write block of the multiple write blocks to a distinct storage node of a plurality of storage nodes;
• responsive to writing the target data, in the token pool, converting available-tokens used to write the target data to unavailable-tokens;
• receiving acknowledgement that the writing of the target data is complete;
• and
• responsive to receiving the acknowledgement, converting the unavailable-tokens used to write the target data to available-tokens.

One or more non-transitory computer-readable storage media storing instructions which, when executed by one or more computing devices, cause:

• performing multiple writes for a host device of a plurality of host devices;
• wherein each write of the multiple writes is to one of multiple storage nodes;
• wherein the multiple writes include writes to a plurality of the multiple storage nodes;
• deriving a host-level latency for the host device based on write-level latencies experienced during the writes to the plurality of storage nodes;
• receiving one or more write requests from a guest workload running on the host device, wherein the plurality of write requests comprise target data;
• delaying transmission of the target data from the host device for a period of time determined based, at least in part, on the host-level latency;
• after the period of time, transmitting the target data from the host device.

One or more non-transitory computer-readable storage media storing instructions which, when executed by one or more computing devices, cause:

• receiving a write request from a guest process running on a host device, wherein the write request comprises target data;
• determining whether a token pool of the host device has sufficient available-tokens to write the target data;
• when there are not sufficient available-tokens in the token pool of the host device to write the target data, postponing writing the target data until there are enough available-tokens in the token pool of the host device to write the target data;
• when there are sufficient available-tokens in the token pool of the host device to write the target data:
• breaking the target data into multiple write blocks;
• writing each write block of the multiple write blocks to a distinct storage node of a plurality of storage nodes;
• responsive to writing the target data, in the token pool, converting available-tokens used to write the target data to unavailable-tokens;
• receiving acknowledgement that the writing of the target data is complete;
• and
• responsive to receiving the acknowledgement, converting the unavailable-tokens used to write the target data to available-tokens.