1c
Persistent Memory:
Accessed as Memory,
Managed as Storage
Fellow, SanDisk
August 7, 2014
Nisha Talagala
2
Forward-Looking Statements
During our meeting today we may make forward-looking statements.
Any statement that refers to...
3
Hybrid Memory-Storage Hierarchy
August 7, 2014
L1, L2, L3 CPU Caches DRAM Persistent
Memories
Flash Hard Drive
Microseco...
4
ACCESSED WITH THE
SPEED AND SIMPLICITY OF MEMORY
MANAGED WITH THE
FAMILIARITY AND CAPACITY OF STORAGE
5
NVMFS (Non-Volatile Memory File System)
August 7, 2014
 Appears as Linux file system
– Provides performance to applicat...
6
NVMFS: Consistent Low Latency
Sample
Latency(µs) NVMFS vs XFS write latency
1000 512 Byte Sequential Write with O_DIRECT...
7
ACM (Auto Commit Memory): A Persistent Memory
August 7, 2014
 Persistent, granular, byte and cacheline size updates
– L...
8
An example use case:
Persistent Memory Interface:
Software providing access to persistent memory
August 7, 2014
Auto-Com...
9
Example Use Case: MySQL Database
August 7, 2014
8,000
16,000 15,750
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,00...
10
Appendix
Upcoming SlideShare
Loading in …5
×

Persistent Memory: Accessed as Memory, Managed as Storage by Nisha Talagala

1,001 views
901 views

Published on

SanDisk Fellow Nisha Talagala discusses coming changes to the memory-storage hierarchy, where media will be accessed like memory but managed like storage.

Published in: Technology
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,001
On SlideShare
0
From Embeds
0
Number of Embeds
45
Actions
Shares
0
Downloads
12
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

  • This work is on the forefront of the coming changes to the memory-storage hierarchy, where media will be accessed like memory and managed like storage. Memory suppliers are working hard to supply a whole family of persistent memory mediums for companies like Fusion-io to build next generation systems.

    This chart demonstrates combining Persistent Memories with Flash into a hybrid memory-storage hierarchy. Fusion-io is doing this through our Auto-Commit Memory software. In our demonstration, we write to the ioDrive2 like memory at near-DRAM speeds, BUT WITH PERSISTENCE, while managing the data like storage through common file system mechanisms.

    In the chart on this slide, note the difference in access latencies to the various media, depicted in CPU cycles. This highlights the obvious benefit of a hybrid memory-storage hierarchy, where the speed and simplicity of Persistent Memory is transparently bonded with the capacity and familiarity of flash storage.
  • On March 9, Fusion-io announced the ability to deliver more than 9 million I/O operations per second (IOPS) from a single ioDrive2.
     
    This was done using a new memory-storage hierarchy. This hierarchy provides the speed and simplicity of persistent memory with the familiarity and capacity of storage. Through our Auto Commit Memory software, we have transparently bonded persistent memory with flash memory, and have wrapped it all in a POSIX-standard file system.

    We have begun applying technologies to revolutionize common industry workloads, such as database transaction logging.
  • As you can see, directFS also achieves consistent low latency compared to XFS, as you can see on that blue line.
  • With native APIs flash can deliver more functionality without inhibiting performance.
  • directFS: Nonvolatile memory optimized file system eliminates duplicate work between the host file system and flash memory management software
    Auto-Commit Memory: Maintains the persistence of flash at performance levels in the nanoseconds, running on directFS
    TLOG Interface: Library providing memory-speed transaction logging with immediate persistence, with the convenience of file system management.
  • NISHA – INVITE DAVID BACK ON STAGE TO WRAP UP
  • Persistent Memory: Accessed as Memory, Managed as Storage by Nisha Talagala

    1. 1. 1c Persistent Memory: Accessed as Memory, Managed as Storage Fellow, SanDisk August 7, 2014 Nisha Talagala
    2. 2. 2 Forward-Looking Statements During our meeting today we may make forward-looking statements. Any statement that refers to expectations, projections or other characterizations of future events or circumstances is a forward-looking statement, including those relating to industry trends, future memory technology and product capabilities and performance. Information in this presentation may also include or be based upon information from third parties, which reflects their expectations and projections as of the date of issuance. Actual results may differ materially from those expressed in these forward-looking statements due to factors detailed under the caption “Risk Factors” and elsewhere in the documents we file from time to time with the SEC, including our annual and quarterly reports. We undertake no obligation to update these forward-looking statements, which speak only as of the date hereof or as of the date of issuance by a third party, as the case may be.
    3. 3. 3 Hybrid Memory-Storage Hierarchy August 7, 2014 L1, L2, L3 CPU Caches DRAM Persistent Memories Flash Hard Drive MicrosecondsNanoseconds CYCLES TO WAIT Hybrid Memory Storage Hierarchy Main Memory System High Performance Disk System Accessed Like Memory and Managed Like Storage Milliseco nds ACCESS DELAY 2 cycles 4 million cycles
    4. 4. 4 ACCESSED WITH THE SPEED AND SIMPLICITY OF MEMORY MANAGED WITH THE FAMILIARITY AND CAPACITY OF STORAGE
    5. 5. 5 NVMFS (Non-Volatile Memory File System) August 7, 2014  Appears as Linux file system – Provides performance to applications “as is” – Focuses only on file namespace – POSIX (Portable Operating System Interface) compatible  Flash – Employs existing flash translation layer for: – Large virtualized addressed space – Direct flash access – Crash recovery mechanisms – Exports primitives through file namespace  Persistent Memory: Application access through filesystem or straight to device
    6. 6. 6 NVMFS: Consistent Low Latency Sample Latency(µs) NVMFS vs XFS write latency 1000 512 Byte Sequential Write with O_DIRECT NVMFS XFS (Extended File System) August 7, 2014
    7. 7. 7 ACM (Auto Commit Memory): A Persistent Memory August 7, 2014  Persistent, granular, byte and cacheline size updates – Latency reductions 10x-20x* – 32x-64x less data written to media  Direct memory access reduces CPU overhead  Integrated with NVMFS for use with regular files  Automatic tiering across persistent memory and flash  Application integrated Storage Class Memory *compared to write operations to an ioMemory device though classic block level interfaces
    8. 8. 8 An example use case: Persistent Memory Interface: Software providing access to persistent memory August 7, 2014 Auto-Commit Memory: Persistent, byte-addressable memory management NVMFS: File system native to ACM and flash memory
    9. 9. 9 Example Use Case: MySQL Database August 7, 2014 8,000 16,000 15,750 0 2,000 4,000 6,000 8,000 10,000 12,000 14,000 16,000 18,000 B A S E L I N E S C E N A R I O 1 S C E N A R I O 2 INNOSIMOPS/SEC Nearly as fast as disabling the transaction log completely. Log transaction through block I/O No Logging Log to ACM
    10. 10. 10 Appendix

    ×