Multilevel multipath widely distributed computational node scenarios

Patent metadata

Field	Value
Publication number	US-9634956-B2
Application number	US-201615066915-A
Country	US
Kind code	B2
Filing date	Mar 10, 2016
Priority date	Feb 12, 2011
Publication date	Apr 25, 2017
Grant date	Apr 25, 2017

Abstract

An application combines on-site and cloud-based calculations. The application selects a cloud computing broker from a list sorted by load. The broker receives computation requests from the application, sends corresponding load-balanced requests to local application service hosts and cloud proxies, receives responses, and sends the responses to the application without identifying the response source as local versus non-local. Each proxy has a logical-physical mapping of cloud-resident worker nodes. A scheduler receives session requests from applications, allocates local and cloud computing resources to a session, and tracks resource usage by session. An application previously run only on-premises can be executed partially on cloud worker nodes without modification. The proxy re-routes unfulfilled requests from a failed worker, undetected by the broker. If all workers assigned to that proxy fail, then the broker reassigns unfulfilled requests to another proxy, undetected by the application.

First claim

What is claimed is:

1. A multilevel multipath widely distributed computational node system comprising:
at least one broker node having an application request-response interface, at least one application service host interface, and at least one wide distribution proxy interface, the broker node also having a logical processor and a memory in operable communication with the logical processor; and
a premises transparency code residing in the memory and having a request store, the premises transparency code configured to receive requests into the request store through the application request-response interface, to send corresponding computation requests to local nodes through the application service host interface(s) and to non-local nodes through the wide distribution proxy interface(s) based on a prior allocation of local and widely distributed node computational resources by a node other than the broker node, to receive computation responses from local nodes through the application service host interface(s) and from non-local nodes through the wide distribution proxy interface(s), and to send the computation responses through the application request-response interface without thereby identifying the source of the computation responses as being from a local node versus a non-local node.

2. The system of claim 1 , wherein the system further comprises at least one wide distribution proxy node having a logical processor and a memory in operable communication with the logical processor, a wide distribution proxy code residing in the memory and having a broker premises transparency code interface, a plurality of worker node interfaces, and a logical-physical worker node mapping of widely distributed worker nodes, the wide distribution proxy code configured to receive computation requests through the broker premises transparency code interface, to send the computation requests through the worker node interfaces using the mapping, to receive computation responses through the worker node interfaces, and to send the computation responses through the broker premises transparency code interface.

3. The system of claim 1 , wherein the system comprises a plurality of such broker nodes, each node having a load, and the system also comprises a list of broker nodes, the list residing in the memory and reciting the broker nodes in order according to their respective loads.

4. The system of claim 1 , wherein the system further comprises a head node having a logical processor and a memory in operable communication with the logical processor, an application session interface, and a scheduler code residing in the memory, the scheduler code configured to receive session requests through the application session interface, to allocate resources to a session, and to track resource usage by a session, without maintaining a list of worker node physical identifiers.

5. The system of claim 1 , wherein the system comprises a plurality of such broker nodes, each node having a load, and the system further comprises:
a head node having a logical processor and a memory in operable communication with the logical processor, an application session interface, and a scheduler code residing in the memory, the scheduler code configured to receive session requests through the application session interface, to allocate resources to a session, and to track resource usage by a session, a list of broker nodes also residing in the memory and reciting the broker nodes in order according to their respective loads; and
at least one wide distribution proxy node having a logical processor and a memory in operable communication with the logical processor, a wide distribution proxy code residing in the memory and having a broker premises transparency code interface, a plurality of worker node interfaces, and a logical-physical worker node mapping of widely distributed worker nodes, the wide distribution proxy code configured to receive computation requests through the broker premises transparency code interface, to send the computation requests through the worker node interfaces using the mapping, to receive computation responses through the worker node interfaces, and to send the computation responses through the broker premises transparency code interface.

6. A process for multilevel multipath widely distributed computational node management, the process comprising the steps of:
a broker node receiving multiple requests from at least one application node;
the broker node sending corresponding computation requests to at least one wide distribution proxy node and at least one local node based on a prior allocation of local and widely distributed node computational resources;
the broker node receiving computation responses from at least one wide distribution proxy node and at least one local node;
the broker node sending the computation responses to at least one application node without thereby identifying the source of the computation responses as being from a local node versus a non-local node.

7. The process of claim 6 , further comprising the broker node load-balancing the computation requests sent to the wide distribution proxy node(s) and the local node(s), thereby distributing computational workload among those nodes according to at least one predetermined load-balancing criterion.

8. The process of claim 6 , further comprising connecting the broker node to a head node, the head node having a scheduler, and the broker node reporting resource usage to the scheduler.

9. The process of claim 6 , further comprising the application node getting a list of broker nodes in the order of least loaded broker node first.

10. The process of claim 6 , further comprising a scheduler in a head node allocating resources for a session and launching the resources.

11. The process of claim 6 , further comprising the wide distribution proxy node(s) maintaining a mapping of logical widely distributed worker nodes to physical widely distributed worker nodes.

12. The process of claim 11 , further comprising the wide distribution proxy node monitoring entries of the mapping for heartbeats, and marking an entry null after a given widely distributed worker node misses a specified number of heartbeats.

13. The process of claim 6 , further comprising the wide distribution proxy node(s) receiving computation requests from the broker node, sending the computation requests to widely distributed worker nodes, receiving computation responses from the widely distributed worker nodes, and sending the computation responses to the broker node without thereby identifying any particular widely distributed worker node as the source of a given computation response.

14. A computer-readable distributed storage medium configured with data and with instructions that when executed by processors cause the processors to perform a process for process for multilevel multipath widely distributed computational node management, the process comprising the steps of:
a broker node sending computation requests to at least one wide distribution proxy node, and to at least one local node, said distribution proxy and local nodes previously allocated by a head node;
the wide distribution proxy node(s) receiving computation requests from the broker node, sending the computation requests to widely distributed worker nodes, receiving computation responses from the widely distributed worker nodes, and sending the computation responses to the broker node without thereby identifying any particular widely distributed worker node as the source of a given computation response;
the broker node receiving computation responses from the wide distribution proxy node(s)

Assignees

• Microsoft Technology Licensing Llc
• Microsoft Technology Licensing Llc

Inventors

• Xu Mingqiang
• Su Jun
• Burgess Gregory M
• Ma Tianchi
• Alam Salim
• Ciapala Rich

Classifications

• H04L47/70 Primary

  Admission control; Resource allocation · CPC title

• G06F9/505 Primary

  considering the load · CPC title

• G06F2209/5015

  Service provider selection · CPC title

• H04L67/28

  Electricity · mapped topic

• H04L67/1097

  for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS] · CPC title