Distributed flexible scheduler for converged traffic
A distributed flexible scheduler that dynamically balances network and storage traffic across links is proposed. The scheduler takes into account the bandwidth requirements of workloads provisioned in a cluster and dynamically distributes the network traffic and the storage traffic accordingly. There are three schemes involved in the proposed distributed flexible scheduler. In a first approach of Equal Distribution, network and storage traffic is distributed evenly across the links. In a second approach of Storage Preferred distribution, the aggregate storage bandwidth requirements of workloads exceed the network bandwidth requirements. In a third approach of Network Preferred distribution, the aggregate network bandwidth requirements of workloads exceed the storage bandwidth requirements.
1. A method, comprising:
receiving a first type of traffic from a first plurality of connectors coupled to a first plurality of links, wherein the first type of traffic is a storage type of traffic;
receiving a second type of traffic from a second plurality of connectors coupled to a second plurality of links, wherein the second type of traffic is a network type of traffic;
determining a distribution preference based on a first provisioned bandwidth of the first type of traffic and a second provisioned bandwidth of the second type of traffic; and
outputting the first type of traffic onto a first subset of a plurality of converged links, and outputting the second type of traffic onto a second subset of the plurality of converged links, wherein the first subset and the second subset are determined based on the distribution preference, wherein a storage-preferred distribution is determined when the first provisioned bandwidth exceeds half of a total bandwidth of the converged links, and wherein at least one converged link is reserved for the network type of traffic.
2. The method of claim 1 , wherein the first plurality of links is a storage link comprising a Non-Volatile Memory express (NVMe) over Ethernet link.
3. The method of claim 1 , wherein the second plurality of links is a network link comprising an Ethernet link.
4. The method of claim 1 , wherein at least one of the converged links fails, wherein the first type of traffic and the second type of traffic are pinned to the remaining converged links based on the distribution preference.
5. A method, comprising:
receiving a first type of traffic from a first plurality of connectors coupled to a first plurality of links, wherein the first type of traffic is a storage type of traffic;
receiving a second type of traffic from a second plurality of connectors coupled to a second plurality of links, wherein the second type of traffic is a network type of traffic;
determining a distribution preference based on a first provisioned bandwidth of the first type of traffic and a second provisioned bandwidth of the second type of traffic; and
outputting the first type of traffic onto a first subset of a plurality of converged links, and outputting the second type of traffic onto a second subset of the plurality of converged links, wherein the first subset and the second subset are determined based on the distribution preference, wherein a network-preferred distribution is determined when the second provisioned bandwidth exceeds half of a total bandwidth of the converged links, and wherein at least one converged link is reserved for the storage type of traffic.
6. The method of claim 5 , wherein the first plurality of links is a storage link comprising a Non-Volatile Memory express (NVMe) over Ethernet link.
7. The method of claim 5 , wherein the second plurality of links is a network link comprising an Ethernet link.
8. The method of claim 5 , wherein at least one of the converged links fails, wherein the first type of traffic and the second type of traffic are pinned to the remaining converged links based on the distribution preference.
9. A converged IO controller, comprising:
a first plurality of connectors coupled to a first plurality of links for receiving a first type of traffic that is a storage type of traffic;
a second plurality of connectors coupled to a second plurality of links for receiving a second type of traffic that is a network type of traffic;
a plurality of converged connectors coupled to a plurality of converged links; and
a scheduler that determines a distribution preference based on a first provisioned bandwidth of the first type of traffic and a second provisioned bandwidth of the second type of traffic, wherein the first type of traffic is pinned to a first subset of a plurality of converged links, and the second type of traffic is pinned to a second subset of the plurality of converged links, and wherein the first subset and the second subset are determined based on the distribution preference, wherein a storage-preferred distribution is determined when the first provisioned bandwidth exceeds half of a total bandwidth of the converged links, and wherein at least one converged link is reserved for the network type of traffic.
10. The controller of claim 9 , wherein the first plurality of links is a storage link comprising a Non-Volatile Memory express (NVMe) over Ethernet link.
11. The controller of claim 9 , wherein the second plurality of links is a network link comprising an Ethernet link.
12. The controller of claim 9 , wherein at least one of the converged links fails, wherein the first type of traffic and the second type of traffic are pinned to the remaining converged links based on the distribution preference.
13. A converged IO controller, comprising:
a first plurality of connectors coupled to a first plurality of links for receiving a first type of traffic that is a storage type of traffic;
a second plurality of connectors coupled to a second plurality of links for receiving a second type of traffic that is a network type of traffic;
a plurality of converged connectors coupled to a plurality of converged links; and
a scheduler that determines a distribution preference based on a first provisioned bandwidth of the first type of traffic and a second provisioned bandwidth of the second type of traffic, wherein the first type of traffic is pinned to a first subset of a plurality of converged links, and the second type of traffic is pinned to a second subset of the plurality of converged links, and wherein the first subset and the second subset are determined based on the distribution preference, wherein a network-preferred distribution is determined when the second provisioned bandwidth exceeds half of a total bandwidth of the converged links, and wherein at least one converged link is reserved for the storage type of traffic.
14. The controller of claim 13 , wherein the first plurality of links is a storage link comprising a Non-Volatile Memory express (NVMe) over Ethernet link.
15. The controller of claim 13 , wherein the second plurality of links is a network link comprising an Ethernet link.
16. The controller of claim 13 , wherein at least one of the converged links fails, wherein the first type of traffic and the second type of traffic are pinned to the remaining converged links based on the distribution preference.