Embed Size (px)
DESCRIPTION
USE CASE – Local NVDIMM – Kernel Access -Memory Side Cache -Hierarchical Tiered Memory Client Workstation or Enterprise Server On Package High-Bandwidth Memory Hierarchical Tiered Memory and Caching -On-Package High-Bandwidth Memory caching DDR4 main memory -DDRT4 Main Memory caching byte-addressable NVDIMM -NVDIMM caching SSDs -Typical OS implementations will require extensive kernel access DDR4 Main Memory Byte addressable, direct access, NVDIMM SSD Asynchronous Demand On Package High-Bandwidth Memory Cache Requests Asynchronous Demand On Package High-Bandwidth Memory Cache Completions Asynchronous Demand DDR4 Main Memory Cache Requests Asynchronous Demand DDR4 Main Memory Cache Completions Asynchronous Demand Byte-Addressable NVDIMM Cache Requests (Writes, Flushes, Commits) Asynchronous Demand Byte-Addressable NVDIMM Cache Completions Asynchronous Demand SSD Requests Asynchronous Demand SSD Completions
Citation preview
Local and Remote byte-addressable NVDIMM High-level Use CasesChet Douglas – Principal NVDIMM SW Architect - DCG09-11-15
USE CASE – Local NVDIMM – Kernel Access-Memory Side Cache-Hierarchical Tiered Memory
Asynchronous Demand DDR4 Main Memory Cache Requests
Asynchronous Demand Byte-Addressable NVDIMM Cache Requests (Writes, Flushes, Commits)Asynchronous Demand DDR4 Main Memory Cache Completions
Asynchronous Demand On Package High-Bandwidth Memory Cache RequestsAsynchronous Demand On Package High-Bandwidth Memory Cache Completions
Client Workstationor Enterprise Server
Byte addressable,direct access,NVDIMM
Hierarchical Tiered Memory and Caching-On-Package High-Bandwidth Memory caching DDR4 main memory-DDRT4 Main Memory caching byte-addressable NVDIMM-NVDIMM caching SSDs-Typical OS implementations will require extensive kernel access
DDR4 Main Memory
On Package High-Bandwidth Memory
SSD
Asynchronous Demand Byte-Addressable NVDIMM Cache CompletionsAsynchronous Demand SSD RequestsAsynchronous Demand SSD Completions
USE CASE – Local NVDIMM – Kernel Access-SW RAID – Close “R5 Write Hole”-SW RAID – Cache for SSD RAID Volumes
Asynchronous Demand SW RAID Data Vulnerability Cache Requests
Asynchronous Demand SW RAID RequestsAsynchronous Demand SW RAID Data Vulnerability Cache Completions
Asynchronous Demand SW RAID Acceleration Cache RequestsAsynchronous Demand SW RAID Acceleration Cache Completions
Client Workstationor Enterprise Server
Byte addressable,direct access,NVDIMM
SW RAID Use Cases – Kernel Access-Typically SW RAID products utilize OS Kernel RAID Storage Drivers and UEFI RAID Pre-Boot drivers to implement SW RAID functionality.-Performance acceleration caches or caches to close vulnerabilities in RAID Data replication, require shared persistent kernel access to regions of the NVDIMM by both UEFI RAID Pre-Boot drivers and OS Kernel RAID Storage Drivers
SSD
Asynchronous Demand SW RAID Completions
SSD RAID Volume
Byte addressable,direct access,NVDIMM
SSD RAID Volume Acceleration Cache
SSD RAID-5 Volume “Write Hole”Temp Data Store
USE CASE – Local & Remote NVDIMMPublic & Private Cloud-Synchronous (1:1) (active:passive) -Database Replication
Asynchronous Database Replication - RDMA Write
ActiveServerNode
ActiveDatabase
Master
Client
Synchronous Database Replication SNIA Optimized Flush
Synchronous Database Replication SNIA Optimized Flush Completion
Byte addressable,direct access,
NVDIMM
Byte addressable,direct access,
NVDIMM
Database copies kept in lock step
Asynchronous Demand Database Master Updates
Asynchronous Demand Database Reads
Synchronous Passive
DatabaseSlave
PUBLIC & PRIVATE CLOUD
ActiveServer
Node
USE CASE – Local & Remote NVDIMMPublic & Private Cloud-Asynchronous (1:1) (active:passive) -Database Replication
Asynchronous Demand Database Master Updates
ActiveServerNode
ActiveDatabase
MasterAsynchronous
PassiveDatabase
Slave
Client
Synchronous Database Logfile Replication SNIA Optimized FlushSynchronous Database Logfile Replication SNIA Optimized Flush Completion
Byte addressable,direct access,
NVDIMM
Byte addressable,direct access,
NVDIMM
Database copies updated asynchronously as logfile updates are pushed to slave nodes
Local Database Update due to Logfile update
Local Logfile Update due to Database update
Database Master Logfile
Database Slave Logfile
Asynchronous Database Logfile Replication - RDMA Write
Asynchronous Demand Database Reads
PUBLIC & PRIVATE CLOUD
ActiveServerNode
USE CASE – Remote NVDIMMPublic & Private Cloud-Converged Disaggregated Network Topology-SW Defined Storage (SDS)
Asynchronous Storage Updates - RDMA Write
Compute Node
Compute NodeVirtual Clients
(IaaS/PaaS)
Synchronous Storage Updates SNIA Optimized Flush
Synchronous Storage Updates SNIA Optimized Flush Completion
Byte addressable,direct access,
NVDIMM
Asynchronous Demand SDS Updates
Asynchronous Demand SDS Reads
DirectAttachedNVDIMM
PUBLIC & PRIVATE CLOUDStorage
ProcessingNode
Converged Disaggregated Network Topology Compute Nodes and Storage ProcessingNodes are physically separate nodes
USE CASE – Remote NVDIMMPublic & Private Cloud-Hyper-Converged Network Topology-SW Defined Storage (SDS)
Asynchronous Storage Updates - RDMA Write
Compute Node
Compute NodeVirtual Clients
(IaaS/PaaS)
Synchronous Storage Updates SNIA Optimized Flush
Synchronous Storage Updates SNIA Optimized Flush Completion
Byte addressable,direct access,
NVDIMMAsynchronous Demand SDS Updates
Asynchronous Demand SDS Reads
Compute Node Direct AttachedNVDIMM
PUBLIC & PRIVATE CLOUD
VirtualStorageProcessingNode
Hyper-Converged Network Topology Compute Node and Storage Processing Node are in the same physically node
VirtualNetwork
USE CASE – Remote NVDIMMEnterprise Appliance
Asynchronous Appliance Updates - RDMA Write
Synchronous Appliance Update SNIA Optimized Flush
Synchronous Appliance Update SNIA Optimized Flush Completion
Asynchronous Storage Updates
Enterprise ServerCompute Node
Byte addressable,direct access,
NVDIMM
Network Attached Enterprise Appliance Direct AttachedNVDIMM
Enterprise ServerStorage Processing Node
Byte addressable,direct access,
NVDIMM
Storage Attached Enterprise Appliance w Internal Network Fabric Switch & Direct Attached NVDIMM
USE CASE – Remote NVDIMMHPC – PGAS, SHMEM, MPI, Human Brain Project-Distributed (1:N) (active:active) Database
Asynchronous Database Updates - RDMA Write
Synchronous DistributedDatabase
Client
Asynchronous Database Replication SNIA Optimized FlushAsynchronous Database Replication SNIA Optimized Flush Completion
Database is distributed across many nodes all actively taking part in Database references and updatesByte addressable,
direct access,NVDIMMAsynchronous Demand Parallel Database Reads
Distributed Database Reference
Asynchronous Demand Database Updates
Synchronous DistributedDatabase
Synchronous DistributedDatabase
Synchronous DistributedDatabase
ActiveServerNode
ActiveServerNode
ActiveServerNode
ActiveServerNode
Byte addressable,direct access,
NVDIMM
Byte addressable,direct access,
NVDIMM
Byte addressable,direct access,
NVDIMM
IO typically <= 256Bytes, Key/Value pair or pointer updates
USE CASE – Remote NVDIMMHA & HPC-Synchronous (1:N) (active:passive) -Database Replication w Parallel Access
ActiveServer SlaveNode
Asynchronous Database Updates - RDMA Write
ActiveServerNode
DatabaseMaster
Synchronous Database
SlaveClient
Synchronous Database Replication SNIA Optimized Flush
Synchronous Database Replication SNIA Optimized Flush Completion
Byte addressable,direct access,
NVDIMM
Byte addressable,direct access,
NVDIMM
High Availability – Database copies kept in lock step
ActiveServerSlaveNode
Synchronous Database
Slave
Byte addressable,direct access,
NVDIMM
ActiveServerSlaveNode
Synchronous Database
Slave
Byte addressable,direct access,
NVDIMM
Asynchronous Demand Parallel Database Reads - RDMA Read
High Performance Computing - Parallel Database Reads typically load-balanced across all Passive Server Nodes
Asynchronous Demand Database Master Updates
![Homework 9: Software Design Considerations · The PIC24F series has a separate 16-bit wide data memory space, divided into byte addressable blocks (8 bits) [1]. Byte addresses 0x0000](https://img.pdfslide.net/doc/110x75/5e72aa6435ceab51450255e9/homework-9-software-design-considerations-the-pic24f-series-has-a-separate-16-bit.jpg)
