# US 2015/0103837 A1 · Leveraging hardware accelerators for scalable distributed stream processing in a network environment

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2015/0103837 A1
- Document ID: US20150103837A1
- Status: Published application
- Kind: A1
- Application number: 14/054,542
- Filed: 2013-10-15
- Published: 2015-04-16
- Earliest claimed: 2013-10-15
- Assignees: Cisco Technology, Inc.
- Applicants: Cisco Technology, Inc.
- Inventors: Debojyoti Dutta
- Web page: https://patentgrep.com/patent/US20150103837A1

## CPC

- H04L 47/125: by balancing the load, e.g. traffic engineering
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)

## Abstract

An example method for leveraging hardware accelerators for scalable distributed stream processing in a network environment is provided and includes allocating a plurality of hardware accelerators to a corresponding plurality of bolts of a distributed stream in a network, facilitating a handshake between the hardware accelerators and the corresponding bolts to allow the hardware accelerators to execute respective processing logic according to the corresponding bolts, and performing elastic allocation of hardware accelerators and load balancing of stream processing in the network. The distributed stream comprises a topology of at least one spout and the plurality of bolts. In specific embodiments, the allocating includes receiving capability information from the bolts and the hardware accelerators, and mapping the hardware accelerators to the bolts based on the capability information. In some embodiments, facilitating the handshake includes executing a shadow process to interface between the hardware accelerator and the distributed stream.

## Claims

### Claim 1

A method, comprising:
allocating a plurality of hardware accelerators to a corresponding plurality of bolts of a distributed stream in a network, wherein the distributed stream comprises a topology of at least one spout and the plurality of bolts;
facilitating a handshake between the hardware accelerators and the corresponding bolts to allow the hardware accelerators to execute respective processing logic on data elements of the distributed stream according to the corresponding bolts; and
performing elastic allocation of the hardware accelerators and load balancing of stream processing in the network.

### Claim 2 (depends on 1)

The method of claim 1, wherein allocating comprises:
receiving capability information from the bolts and the hardware accelerators; and
mapping the hardware accelerators to the bolts based on the capability information.

### Claim 3 (depends on 2)

The method of claim 2, wherein the capability information from the bolts comprises respective locations in the distributed stream and identities of the bolts, wherein the capability information from the hardware accelerators comprises respective network locations and capabilities of the hardware accelerators.

### Claim 4 (depends on 1)

The method of claim 1, wherein facilitating the handshake comprises executing a shadow process to interface between the hardware accelerators and the corresponding bolts.

### Claim 5 (depends on 1)

The method of claim 1, wherein the handshake comprises an exchange of the capability information and processing information between the hardware accelerators and the corresponding bolts.

### Claim 6 (depends on 5)

The method of claim 5, wherein the processing information comprises:
logic to extract input data from the distributed stream for processing, format input data, and format output data to the distributed stream after processing; and
destination of the output data in the network.

### Claim 7 (depends on 1)

The method of claim 1, wherein performing elastic allocation of the hardware accelerators and load balancing of stream processing comprises:
monitoring hardware accelerator workload in the network; and
instructing at least one bolt to spawn itself and adding a new hardware accelerator based on the workload.

### Claim 8 (depends on 7)

The method of claim 7, further comprising:
receiving capability information from the spawned bolt and the new hardware accelerator;
mapping the spawned bolt to the new hardware accelerator; and
facilitating the handshake between the spawned bolt and the new hardware accelerator.

### Claim 9 (depends on 7)

The method of claim 7, wherein, if the workload indicates that one of the hardware accelerators is over-loaded, transferring at least a portion of stream processing to another one of the hardware accelerators.

### Claim 10 (depends on 7)

The method of claim 7, wherein if the workload indicates that some of the hardware accelerators are not loaded, decommissioning the unloaded hardware accelerators and updating the mapping.

### Claim 11

Non-transitory tangible media that includes instructions for execution, which when executed by a processor, is operable to perform operations comprising:
allocating a plurality of hardware accelerators to a corresponding plurality of bolts of a distributed stream in a network, wherein the distributed stream comprises a topology of at least one spout and the plurality of bolts;
facilitating a handshake between the hardware accelerators and the corresponding bolts to allow the hardware accelerators to execute respective processing logic on data elements of the distributed stream according to the corresponding bolts; and
performing elastic allocation of the hardware accelerators and load balancing of stream processing in the network.

### Claim 12 (depends on 11)

The media of claim 11, wherein allocating comprises:
receiving capability information from the bolts and the hardware accelerators; and
mapping the hardware accelerators to the bolts based on the capability information.

### Claim 13 (depends on 11)

The media of claim 11, wherein facilitating the handshake comprises executing a shadow process to interface between the hardware accelerators and the corresponding bolts.

### Claim 14 (depends on 11)

The media of claim 11, wherein the handshake comprises an exchange of the capability information and processing information between the bolts and corresponding hardware accelerators.

### Claim 15 (depends on 11)

The media of claim 11, wherein performing elastic allocation of the hardware accelerators and load balancing of stream processing comprises:
monitoring hardware accelerator workload in the network; and
instructing at least one bolt to spawn itself and adding a new hardware accelerator based on the workload.

### Claim 16

An apparatus, comprising:
a broker executing in network environment comprising a memory element for storing data and a processor, wherein the processor executes instructions associated with the data, wherein the processor and the memory element cooperate, such that the apparatus is configured for:
allocating a plurality of hardware accelerators to a corresponding plurality of bolts of a distributed stream in a network, wherein the distributed stream comprises a topology of at least one spout and the plurality of bolts;
facilitating a handshake between the hardware accelerators and the corresponding bolts to allow the hardware accelerators to execute respective processing logic on data elements of the distributed stream according to the corresponding bolts; and
performing elastic allocation of the hardware accelerators and load balancing of stream processing in the network.

### Claim 17 (depends on 16)

The apparatus of claim 16, wherein allocating comprises:
receiving capability information from the bolts and the hardware accelerators; and
mapping the hardware accelerators to the bolts based on the capability information.

### Claim 18 (depends on 16)

The apparatus of claim 16, wherein facilitating the handshake comprises executing a shadow process to interface between the hardware accelerators and the corresponding bolts.

### Claim 19 (depends on 16)

The apparatus of claim 16, wherein the handshake comprises an exchange of the capability information and processing information between the bolts and corresponding hardware accelerators.

### Claim 20 (depends on 16)

The apparatus of claim 16, wherein performing elastic allocation of the hardware accelerators and load balancing of stream processing comprises:
monitoring hardware accelerator workload in the network; and
instructing at least one bolt to spawn itself and adding a new hardware accelerator based on the workload.
