High performance raid operations offload with minimized local buffering
The present technique presents a hardware mechanism by which high performance computational engines utilize external/system memory buffers for data source and sync thus requiring a minimized amount of local buffering and imposing almost no buffer or data size limitations.
1. A system for performing an operation on a computer node, comprising:
an operation sequencer to receive an incoming descriptor which includes an opcode for requested operation associated with an XOR operation, and a data source in a system memory of a computer node;
a P buffer shim for reading or writing to a P buffer in the system memory of the computer node;
a Q buffer shim for reading or writing to a Q buffer in the system memory of the computer node; and
computation logic to perform the requested operation associated with the XOR operation by:
reading data from the data source in the system memory of the computer node;
writing data from the data source to the P buffer in the system memory using the P buffer shim;
writing data from the data source to the Q buffer in the system memory using the Q buffer shim;
reading a first value from the P buffer in the system memory using the P buffer shim;
XORing the first value obtained from the P buffer in the system memory with a second value from the data source in system memory to obtain a first result;
writing the first result to the P buffer in the system memory using the P buffer shim;
reading a second value from the Q buffer in the system memory using the Q buffer shim;
multiplying the first result with the second value obtained from the Q buffer in the system memory to obtain a second result; and
writing the second result out to the Q buffer using the Q buffer shim.
2. The system of claim 1 , wherein the P buffer shim includes a P buffer write shim and a P buffer read shim.
3. The system of claim 2 , wherein the Q buffer shim includes a Q buffer write shim and a Q buffer read shim.
4. The system of claim 1 , wherein computation logic creates an outgoing descriptor indicating the requested operation is complete.
5. The system of claim 4 , wherein the outgoing descriptor is sent to an originating node of the incoming descriptor.
6. The system of claim 1 , wherein the system further comprises a coherency buffer including a set of entries and the incoming descriptor includes a state index value and wherein when the state index value matches an entry in the set of entries of the coherency buffer the requested operation is stalled.
7. A method for performing an operation on a computer node, comprising:
receiving an incoming descriptor which includes an opcode for requested operation associated with an XOR operation, and a data source in a system memory of a computer node;
providing a P buffer shim for reading or writing to a P buffer in the system memory of the computer node;
providing a Q buffer shim for reading or writing to a Q buffer in the system memory of the computer node; and
performing the requested operation associated with the XOR operation by:
reading data from the data source in the system memory of the computer node;
writing data from the data source to the P buffer in the system memory using the P buffer shim;
writing data from the data source to the Q buffer in the system memory using the Q buffer shim;
reading a first value from the P buffer in the system memory using the P buffer shim;
XORing the first value obtained from the P buffer in the system memory with a second value from the data source in system memory to obtain a first result;
writing the first result to the P buffer in the system memory using the P buffer shim;
reading a second value from the Q buffer in the system memory using the Q buffer shim;
multiplying the first result with the second value obtained from the Q buffer in the system memory to obtain a second result; and
writing the second result out to the Q buffer using the Q buffer shim.
8. The method of claim 7 , wherein the P buffer shim includes a P buffer write shim and a P buffer read shim.
9. The method of claim 8 , wherein the Q buffer shim includes a Q buffer write shim and a Q buffer read shim.
10. The method of claim 7 , further comprising creating an outgoing descriptor indicating the requested operation is complete.
11. The method of claim 10 , wherein the outgoing descriptor is sent to an originating node of the incoming descriptor.
12. The method of claim 7 , further comprising providing a coherency buffer including a set of entries, wherein the incoming descriptor includes a state index value and when the state index value matches an entry in the set of entries of the coherency buffer the requested operation is stalled.
13. A non-transitory computer readable medium comprising instructions for:
receiving an incoming descriptor which includes an opcode for requested operation associated with an XOR operation, and a data source in a system memory of a computer node;
providing a P buffer shim for reading or writing to a P buffer in the system memory of the computer node;
providing a Q buffer shim for reading or writing to a Q buffer in the system memory of the computer node; and
performing the requested operation associated with the XOR operation by;
reading data from the data source in the system memory of the computer node;
writing data from the data source to the P buffer in the system memory using the P buffer shim;
writing data from the data source to the Q buffer in the system memory using the Q buffer shim;
reading a first value from the P buffer in the system memory using the P buffer shim;
XORing the first value obtained from the P buffer in the system memory with a second value from the data source in system memory to obtain a first result;
writing the first result to the P buffer in the system memory using the P buffer shim;
reading a second value from the Q buffer in the system memory using the Q buffer shim;
multiplying the first result with the second value obtained from the Q buffer in the system memory to obtain a second result; and
writing the second result out to the Q buffer using the Q buffer shim.
14. The non-transitory computer readable medium of claim 13 , wherein the P buffer shim includes a P buffer write shim and a P buffer read shim.
15. The non-transitory computer readable medium of claim 14 , wherein the Q buffer shim includes a Q buffer write shim and a Q buffer read shim.
16. The non-transitory computer readable medium of claim 13 , further comprising creating an outgoing descriptor indicating the requested operation is complete.
17. The non-transitory computer readable medium of claim 16 , wherein the outgoing descriptor is sent to an originating node of the incoming descriptor.
18. The non-transitory computer readable medium of claim 13 , further comprising providing a coherency buffer including a set of entries, wherein the incoming descriptor includes a state index value and when the state index value matches an entry in the set of entries of the coherency buffer the requested operation is stalled.