Remote memory management when switching optically-connected memory

• US 9,104,587 B2
• Filed: 02/06/2013
• Issued: 08/11/2015
• Est. Priority Date: 04/13/2012
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Schenfeld, Eugen, Chakraborty, Abhirup
• Primary Examiner(s): Bragdon, Reginald
• Assistant Examiner(s): VO, THANH DUC
• Application Number:
  US13/760,937
• Publication Number:
  US 20130275705A1
• Time in Patent Office: 916 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 12/023 Free address space management
• CPC Class Codes: G06F 12/0284 Multiple user address space allocation, e.g. using different base addresses interprocessor communication G06F15/163
• CPC Class Codes: G06F 12/0292 using tables or multilevel address translation means G06F12/023 takes precedence; address translation in virtual memory systems G06F12/10
• CPC Class Codes: G06F 12/0891 using clearing, invalidating or resetting means
• CPC Class Codes: G06F 12/10 Address translation
• CPC Class Codes: G06F 12/1009 using page tables, e.g. page table structures
• CPC Class Codes: G06F 12/1027 using associative or pseudo-associative address translation means, e.g. translation look-aside buffer [TLB]
• CPC Class Codes: G06F 13/404 with address mapping
• CPC Class Codes: G06F 13/4068 Electrical coupling
• CPC Class Codes: G06F 13/42 Bus transfer protocol, e.g. handshake; Synchronisation
• CPC Class Codes: G06F 2212/50 Control mechanisms for virtual memory, cache or TLB
• CPC Class Codes: G06F 2212/60 Details of cache memory
• CPC Class Codes: G06F 2212/683 Invalidation
• CPC Class Codes: G06F 3/0604 Improving or facilitating administration, e.g. storage management
• CPC Class Codes: G06F 3/0631 by allocating resources to storage systems
• CPC Class Codes: G06F 3/0647 Migration mechanisms
• CPC Class Codes: G06F 3/067 Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]

Abstract

A remote memory superpage is retained in a remote memory of the memory blade when reading the remote memory super page of the remote memory into a local memory.

Claims

1. In an optically-connected memory (OCM) system, a method for remote memory management, comprising:

a plurality of processors arranged in processor blades and memory in memory blades, with each of the processor blades and the memory blades organized in racks communicatively coupled through an optical-switching fabric, the processor blades and the memory blades remotely separated,

allocating memory from one of the memory blades upon receiving a memory request from the one of the plurality of processors, wherein a circuit is established with the memory for the one of the plurality of processors,

sending a remote machine memory address (RMMA) space of the memory to the one of the plurality of processors,

retaining a remote memory superpage in a remote memory of a memory blade when reading the remote memory superpage of the remote memory into a local memory, wherein performing the retaining upon dynamically switching the memory through the optical-switching fabric using a selected one of a plurality of available communication patterns to transfer the RMMA space in the memory blades from one of the plurality of processors to the alternative one of the plurality of processors in the processor blades without physically copying data in the memory to the plurality of processors,

transferring a physical memory address space from the one of the processors to an alternative one of the plurality of processors,

supplying a set of remote memory superpages by an optical plane, and assimilating the remote memory superpage by grafting the remote memory superpage into a system memory address (SMA) space of the remote processor for creating a mapping within a page table.

2. The method of claim 1, further including performing the retaining on a side of a source node.

3. The method of claim 1, further including writing back the remote memory superpage onto the remote memory while being evicted from the local memory if the remote memory superpage is deleted upon a read operation.

4. The method of claim 1, further including removing the remote memory super page from the local memory if the remote memory superpage is not dirty when retaining the remote memory superpage in the remote memory, wherein back pressure on an optical channel bandwidth is saved.

5. The method of claim 4, further including if the remote memory superpage is dirty, sending only a changed portion of the remote memory superpage to the local memory using a delta coding.

6. The method of claim 1, further including determining whether to retain or remove the remote memory superpage in the remote memory based on an update pattern of the remote memory superpage.

7. The method of claim 1, further including distinguishing ranges of a remote machine memory address (RMMA) space that are overlapping based on one of a channel identification (ID) and a blade ID.

8. The method of claim 1, further including performing:

sending a metadata and a switching request to the one of the plurality of processors,

adjusting the RMMA space and flushing a translation look-aside buffer (TLB),

performing an unmapping operation for removing entries in page tables that map a linear address to the RMMA space,

dynamically switching the memory from an active node of the one of the plurality of processors to a remote active node of the alternative one of the plurality of processors, and

using the page tables by the remote processor for instantaneous access to the data.

9. An optically-connected memory (OCM) system for remote memory management, comprising:

a plurality of nodes, in communication with at least one processor device, including at least a source node, passive nodes, and receiving nodes,

at least one processor blade, at least one memory blade, remotely separated from the at least one processor blade,

an optical plane,

the least one processor device arranged in one of the at least one processor blade and the at least one memory blade and in communication with the optical plane, a translation look-aside buffer (TLB) in communication with the at least one processor device and the optical plane,

at least one memory in the at least one memory blade,

an optical switching fabric communicatively coupled between the at least one processor blade and the at least one memory blade and in communication with the at least one processor device,

the at least one processor device operable in a computing storage environment, wherein the at least one processor device;

allocates memory from one of the memory blades upon receiving a memory request from the one of the plurality of processors, wherein a circuit is established with the memory for the one of the plurality of processors,

sends a remote machine memory address (RMMA) space of the memory to the one of the plurality of processors,

retains a remote memory superpage in a remote memory of a memory blade when reading the remote memory superpage of the remote memory into a local memory,

performs the retaining while dynamically switching the memory through the optical-switching fabric using a selected one of a plurality of available communication patterns to transfer the RMMA space in the memory blades from one of the plurality of processors to an alternative one of the plurality of processors in the processor blades without physically copying data in the memory to the plurality of processors,

transfers a physical memory address space from the one of the processors to the alternative one of the plurality of processors,

supplies a set of remote memory superpages by an optical plane, and assimilates the remote memory superpages by grafting the remote memory superpages into a system memory address (SMA) space of the remote processor for creating a mapping within the page table.

10. The system of claim 9, wherein the at least one processor device performs the collecting, the extracting, and the unmapping on a side of the source node.

11. The system of claim 9, wherein the at least one processor device writes back the remote memory super page onto the remote memory while being evicted from the local memory if the remote memory superpage is deleted upon a read operation.

12. The system of claim 9, wherein the at least one processor device removes the remote memory super page from the local memory if the remote memory super page is not dirty when retaining the remote memory superpage in the remote memory, wherein back pressure on the optical channel bandwidth is saved.

13. The system of claim 12, wherein the at least one processor device, if the remote memory superpage is dirty, sends only a changed portion of the remote memory superpage to the local memory using a delta coding.

14. The system of claim 9, wherein the at least one processor device determines whether to retain or remove the remote memory superpage in the remote memory based on an update pattern of the remote memory superpage.

15. The system of claim 9, wherein the at least one processor device distinguishes ranges of a remote machine memory address (RMMA) space that are overlapping based on one of a channel identification (ID) and a blade ID.

16. The system of claim 9, further including page tables, wherein the at least one processor device:

sends the meta data and a switching request to the one of the plurality of processors,

adjusts a remote machine memory address (RMMA) space and flushing a translation look-aside buffer (TLB),

performs an unmapping operation for removing entries in the page tables that map a linear address to the RMMA space,

dynamically switches the memory from an active node of the one of the plurality of processors to a remote active node of the alternative one of the plurality of processors, and

uses the page tables by the remote processor for instantaneous access to the data.

17. In an optically-connected memory (OCM) system, for remote memory management, a computer program product in a computing environment using a processor device, the computer program product comprising a non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions comprising:

the optically-connected memory (OCM) system includes a plurality of processors arranged in processor blades and memory in memory blades, with each of the processor blades and the memory blades organized in racks communicatively coupled through an optical-switching fabric, the processor blades and the memory blades remotely separated,

a first executable portion that;

allocates the memory from one of the memory blades upon receiving a memory request from the one of the plurality of processors, wherein a circuit is established with the memory for the one of the plurality of processors, a remote machine memory address (RMMA) of the memory sent to the one of the plurality of processors,

retains a remote memory superpage in a remote memory of a memory blade when reading the remote memory superpage of the remote memory into a local memory,

performs the retaining upon dynamically switching the memory through an optical-switching fabric using a selected one of a plurality of available communication patterns to transfer the RMMA space in the memory blades from one of the plurality of processors to an alternative one of the plurality of processors in the processor blades without physically copying data in the memory to the plurality of processors,

supplies a set of remote memory superpages by an optical plane, and assimilates the remote memory superpages by grafting the remote memory superpages into a system memory address (SMA) space of the remote processor for creating a mapping within a page table,

retains a remote memory superpage of the memory in the memory blade when reading the remote memory super page of the memory into the one of the plurality of processors, and

transfers a physical memory address space from the one of the plurality of processors to the alternative one of the plurality of processors.

18. The computer program product of claim 17, further including a fourth executable portion that performs the retaining on a side of a source node.

19. The computer program product of claim 17, further including a second executable portion that writes back the remote memory superpage onto the remote memory while being evicted from the local memory if the remote memory superpage is deleted upon a read operation.

20. The computer program product of claim 17, further including a second executable portion that removes the remote memory superpage from the local memory if the remote memory superpage is not dirty when retaining the remote memory superpage in the remote memory, wherein back pressure on the optical channel bandwidth is saved.

21. The computer program product of claim 20, further including a third executable portion that, if the remote memory superpage is dirty, sends only a changed portion of the remote memory superpage to the local memory using a delta coding.

22. The computer program product of claim 17, further including a second executable portion that determines whether to retain or remove the remote memory superpage in the remote memory based on an update pattern of the remote memory superpage.

23. The computer program product of claim 17, further including a second executable portion that distinguishes ranges of remote machine memory address (RMMA) that are overlapping based on one of a channel identification (ID) and a blade ID.

24. The computer program product of claim 17, further including a second executable portion that:

sends the metadata and a switching request to the one of the plurality of processors,

adjusts a remote machine memory address (RMMA) and flushing a translation look-aside buffer (TLB),

performs an unmapping operation for removing entries in the page tables that map a linear address to the RMMA space,

dynamically switches the memory from an active node of the one of the plurality of processors to a remote active node of the alternative one of the plurality of processors, and

uses the page tables by the remote processor for instantaneous access to the data.