New Features for Oracle Exadata System Software Release 19.x
What's New in Oracle Exadata Database Machine 19.3.0
Persistent Memory Data Accelerator:
Oracle Exadata Storage Server can now use a Persistent Memory (PMEM) Cache in front of Flash Cache. Known as Persistent Memory Data Accelerator, the PMEM cache uses Intel Optane™ DC Persistent Memory Modules (DCPMM). The Database Server uses remote direct memory access (RDMA) to enable 10x faster access latency to remote persistent memory.
Since the persistent memory is used as a shared cache, caching capacity effectively increases by 10x compared to directly using the persistent memory modules as expensive storage. This arrangement makes it cost effective to apply the benefits of persistent memory to multi-terabyte databases.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Database 19c
Oracle Exadata Storage Server X8M-2
Persistent Memory Commit Accelerator
Consistent low latency for redo log writes is critical for OLTP database performance, since transactions are committed only when redo logs are persisted. Furthermore, slow redo log persistence affects critical database algorithms. With the persistent memory commit accelerator, Oracle Database 19c uses Remote Direct Memory Access (RDMA) to write redo records in persistent memory on multiple storage servers. By using RDMA, the redo log writes are up to 8x faster, and excellent resilience is provided because the redo log is persisted on multiple storage servers.
On each storage server, the persistent memory area contains only the recently written log records and persistent memory space is not required for the entire redo log. Therefore, hundreds of databases can share the persistent memory area, enabling consolidation with consistent performance.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Database 19c
Oracle Exadata Storage Server X8M-2
Support for Oracle Exadata X8M Systems:
A new class of Database and Storage Server, designated X8M, are designed to utilize RDMA over Converged Ethernet. Known as RoCE (pronounced rocky), RDMA over Converged Ethernet is the combination of using the existing Remote Direct Memory Access (RDMA) networking verbs, in association with the InfiniBand™ Trade Association (IBTA) standards based extensions for Converged Ethernet to enable high speed fabrics over Ethernet layer 2 and 3 networks. RoCE shares a common API infrastructure with InfiniBand, differing only at the physical and link layers. All existing Oracle Exadata Database Machine high-performance architecture and tuning is compatible with the new network.
Oracle Exadata Database Machine X8M-2 and X8M-8 Database Servers utilize 100Gb/sec RDMA Network cards, which enables higher IO performance, and, when used in conjunction with Oracle Exadata Database Machine X8M-2 Storage Servers with persistent memory acceleration, experience up to 10x lower latency for RDMA reads. All Oracle Exadata Database Machine X8M-2 Storage Servers use the 100Gb/sec RDMA network cards, however the Extended Storage Server (XT) does not receive the persistent memory acceleration. High Capacity and Extreme Flash Storage Servers for the X8M generation receive 1.5 TB of persistent memory per server.
The Oracle Database and Oracle Grid Infrastructure software installations on Oracle Exadata Database Machine X8M database servers must meet the minimum release requirements listed in Oracle Linux Kernel to Unbreakable Enterprise Kernel 5 and Oracle Linux Distribution Upgraded to Oracle Linux 7.7.
Oracle Exadata Database Machine X8M servers must use Oracle Exadata System Software release 19.3.0 or later.
Instant Failure Detection for X8M Systems :
Exadata uses heartbeats to detect the failure of various components. Server failure detection normally requires many missed heartbeats to avoid a false failure diagnosis. Exadata uses remote direct memory access (RDMA) on the RoCE network to quickly confirm server failures. Since RDMA uses hardware, the remote ports respond quickly even when the server is heavily loaded.
Instant failure detection is a unique technology that works transparently and enables incredible availability for OLTP applications.
KVM is the virtualization technology used with Oracle Exadata Database Machine systems configured with RoCE interconnects. KVM provides kernel level hypervisor support, so the host kernel and KVM guest kernel are the same. Oracle Exadata Deployment Assistant (OEDA) is used to create KVM guests and clusters, and a KVM guest uses SR IOV virtual functions for RoCE network connectivity.
Xen is not supported for Oracle Exadata Database Machine systems configured with RoCE interconnects, and KVM is not supported on Oracle Exadata Database Machine systems configured with InfiniBand interconnects.
Faster Encrypted Table Smart Scans
Scans of encrypted tables have been improved by decrypting columns only when needed. Columns in the SELECT set or predicates are the only columns decrypted, not the entire flash cache cacheline. If the first predicate does not return any matches, then the columns for the second predicate are not accessed. This feature has up to 30% faster encrypted smart scans and reduced storage server CPU usage.
For example, consider the following query:
SELECT ename FROM emp
WHERE job LIKE ‘%VP’ AND sal + bonus > 500000
Only the projected columns (ename) and one or more predicated columns are decrypted. The salary in data regions that do not have a VP are not accessed.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Smart In-Memory Columnar Cache with Row IDs
Oracle Exadata System Software evaluates the predicate in a DML statement to determine the rows of interest. These rows are sent back from the storage servers to the database servers, which then uses the ROWIDs to update the rows. This feature expands this functionality to return ROWIDs for data stored in columnar compression format.
For example, consider the following SQL query:
UPDATE weather_history SET temp = (temp – 32) * (5/9)
WHERE country = 'ENGLAND'
Smart scan is used to find the rows that satisfy country = 'ENGLAND' using ROWID values returned from a previous phase of the command processing.
This feature can improve the performance of DML statements by up to 5x.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Database release 18.3.0.0.180717DBRU
Smart In-memory Columnar Cache with Chained Rows
Wide tables, large rows or row migration on update can create chained rows. Smart In-memory Columnar Cache works by creating optimized Single Instruction Multiple Data (SIMD) representation for chained rows when the head and tail pieces fit in the same 1 MB region. This results in up to 3x faster scans for wide tables.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Fast Smart In-memory Columnar Cache Creation
The run-time analyzer finds the best compression algorithm, reusing the dictionary created in-line during the analysis. This provides faster symbol lookup and insert using the new dictionary data structure. The Single Instruction Multiple Data (SIMD) optimized columnar cache creation reuses data from the first pass scan in the flash cache. The result is much faster reads from the flash cache, with an up to 35% improved speed in columnar cache creation.
This feature applies to Exadata Hybrid Columnar Compression tables only.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Update a Single SNMP User Definition
This feature adds new syntax to ALTER CELL and ALTER DBSERVER commands to allow SNMP V3 users to be added, altered, and deleted individually instead of configuring all users in one command string. This features makes it easier to change the password for an individual SNMP user.
ALTER CELL {
| SHUTDOWN SERVICES { RS | MS | CELLSRV | ALL } [IGNORE REDUNDANCY]
| RESTART SERVICES { RS | MS | CELLSRV | ALL } [IGNORE REDUNDANCY]
| RESTART BMC
| STARTUP SERVICES { RS | MS | CELLSRV | ALL }
| LED {ON | OFF}
| DONOTSERVICELED {ON | OFF [FORCE]}
| VALIDATE { MAIL | SNMP | CONFIGURATION }
| VALIDATE SYSLOGCONF selector.node
| CONFIGUREBMC
| BBU { DROP FOR REPLACEMENT | REENABLE }
| { snmpuser=((user_clause)[,(user_clause)[,..]]) | snmpuser.name=(user_clause) }
| attribute_name = attribute_value
[, attribute_name = attribute_value]...
}
ALTER DBSERVER {
| SHUTDOWN SERVICES { RS | MS | ALL }
| RESTART SERVICES { RS | MS | ALL }
| RESTART BMC
| STARTUP SERVICES { RS | MS | ALL }
| LED { ON | OFF }
| VALIDATE { MAIL | SNMP | CONFIGURATION }
| VALIDATE SYSLOGCONF facility.priority
| CONFIGUREBMC
| { snmpuser=((user_clause)[,(user_clause)[,..]]) | snmpuser.name=(user_clause) }
| attribute_name = attribute_value
[, attribute_name = attribute_value]...
}
Securing Storage Server Software Processes with Memory Protection Keys
Memory Protection Keys is a hardware feature found in Oracle Exadata Database Machine X7-2 and newer systems. Memory Protection Keys provide a thread local permission control on memory pages without incurring the high cost of Page Table Entry (PTE) modifications and Translation Look-aside Buffer (TLB) flushes.
The Exadata storage server process that performs block IO (cellsrv) and the processes that perform smart scans (celloflsrv) are now enhanced to run with memory protection keys. This feature is enabled out of the box with no tuning needed. Each thread in these processes needs to obtain access to the appropriate memory protection key before it can access the data. Any access to a piece of memory that does not have the correct key traps the process. This enhances the security and robustness of the storage server processes by eliminating a class of potential memory corruptions.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Exadata Database Machine X7-2
Default XFS File System for X8M Servers
To further improve the security of Oracle Exadata Database Machine, the new X8M servers will use the XFS file system for bare metal, KVM guests, and storage servers.
New partitions are created for /home, /tmp, /var, and /var/log/audit.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Exadata Database Machine X8M servers
Oracle Linux Kernel to Unbreakable Enterprise Kernel 5 and Oracle Linux Distribution Upgraded to Oracle Linux 7.7
This release upgrades Oracle Linux to Unbreakable Enterprise Kernel (UEK) 5, (4.14.35-x.el7uek.x86_64). Oracle Linux Kernel UEK5 introduces support for persistent memory, ROCE, and KVM on Exadata. The Linux distribution is upgraded to Oracle Linux 7.7.
Minimum requirements to upgrade to Oracle Exadata System Software release 19.3.0:
Oracle Grid Infrastructure:
19.4.0.0.0.190716 with patches for 30177140 and 29391849
18.7.0.0.0.190716 with patches for 29637526 and 29391849
12.2.0.1.0.190716 with the patch for 29637526
12.1.0.2.0.190716 with the patch for 30139389
Oracle Database:
19.4.0.0.0.190716 with the patches for 30177140 and 30165493
18.7.0.0.0.190716 with the patch for 30034456
12.2.0.1.0.181016
12.1.0.2.0.180831
11.2.0.4.0.180717
Software Certification Ends for Exadata Database Machine X2 Servers
Software certification ends for Oracle Exadata Database Machine X2 servers. This includes the following servers:
Oracle Exadata Database Machine X2-8 Database Server
Sun Fire X4170 Oracle Database Servers
Sun Fire X4170 M2 Oracle Database Servers
Sun Fire X4800 Oracle Database Servers
Oracle Exadata Storage Server X2-2
Oracle Exadata Storage Server with Sun Fire X4270 M2 Servers
Oracle Exadata Storage Server with Sun Fire X4275 Servers
What's New in Oracle Exadata Database Machine 19.3.0
Persistent Memory Data Accelerator:
Oracle Exadata Storage Server can now use a Persistent Memory (PMEM) Cache in front of Flash Cache. Known as Persistent Memory Data Accelerator, the PMEM cache uses Intel Optane™ DC Persistent Memory Modules (DCPMM). The Database Server uses remote direct memory access (RDMA) to enable 10x faster access latency to remote persistent memory.
Since the persistent memory is used as a shared cache, caching capacity effectively increases by 10x compared to directly using the persistent memory modules as expensive storage. This arrangement makes it cost effective to apply the benefits of persistent memory to multi-terabyte databases.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Database 19c
Oracle Exadata Storage Server X8M-2
Persistent Memory Commit Accelerator
Consistent low latency for redo log writes is critical for OLTP database performance, since transactions are committed only when redo logs are persisted. Furthermore, slow redo log persistence affects critical database algorithms. With the persistent memory commit accelerator, Oracle Database 19c uses Remote Direct Memory Access (RDMA) to write redo records in persistent memory on multiple storage servers. By using RDMA, the redo log writes are up to 8x faster, and excellent resilience is provided because the redo log is persisted on multiple storage servers.
On each storage server, the persistent memory area contains only the recently written log records and persistent memory space is not required for the entire redo log. Therefore, hundreds of databases can share the persistent memory area, enabling consolidation with consistent performance.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Database 19c
Oracle Exadata Storage Server X8M-2
Support for Oracle Exadata X8M Systems:
A new class of Database and Storage Server, designated X8M, are designed to utilize RDMA over Converged Ethernet. Known as RoCE (pronounced rocky), RDMA over Converged Ethernet is the combination of using the existing Remote Direct Memory Access (RDMA) networking verbs, in association with the InfiniBand™ Trade Association (IBTA) standards based extensions for Converged Ethernet to enable high speed fabrics over Ethernet layer 2 and 3 networks. RoCE shares a common API infrastructure with InfiniBand, differing only at the physical and link layers. All existing Oracle Exadata Database Machine high-performance architecture and tuning is compatible with the new network.
Oracle Exadata Database Machine X8M-2 and X8M-8 Database Servers utilize 100Gb/sec RDMA Network cards, which enables higher IO performance, and, when used in conjunction with Oracle Exadata Database Machine X8M-2 Storage Servers with persistent memory acceleration, experience up to 10x lower latency for RDMA reads. All Oracle Exadata Database Machine X8M-2 Storage Servers use the 100Gb/sec RDMA network cards, however the Extended Storage Server (XT) does not receive the persistent memory acceleration. High Capacity and Extreme Flash Storage Servers for the X8M generation receive 1.5 TB of persistent memory per server.
The Oracle Database and Oracle Grid Infrastructure software installations on Oracle Exadata Database Machine X8M database servers must meet the minimum release requirements listed in Oracle Linux Kernel to Unbreakable Enterprise Kernel 5 and Oracle Linux Distribution Upgraded to Oracle Linux 7.7.
Oracle Exadata Database Machine X8M servers must use Oracle Exadata System Software release 19.3.0 or later.
Instant Failure Detection for X8M Systems :
Exadata uses heartbeats to detect the failure of various components. Server failure detection normally requires many missed heartbeats to avoid a false failure diagnosis. Exadata uses remote direct memory access (RDMA) on the RoCE network to quickly confirm server failures. Since RDMA uses hardware, the remote ports respond quickly even when the server is heavily loaded.
Instant failure detection is a unique technology that works transparently and enables incredible availability for OLTP applications.
KVM is the virtualization technology used with Oracle Exadata Database Machine systems configured with RoCE interconnects. KVM provides kernel level hypervisor support, so the host kernel and KVM guest kernel are the same. Oracle Exadata Deployment Assistant (OEDA) is used to create KVM guests and clusters, and a KVM guest uses SR IOV virtual functions for RoCE network connectivity.
Xen is not supported for Oracle Exadata Database Machine systems configured with RoCE interconnects, and KVM is not supported on Oracle Exadata Database Machine systems configured with InfiniBand interconnects.
Faster Encrypted Table Smart Scans
Scans of encrypted tables have been improved by decrypting columns only when needed. Columns in the SELECT set or predicates are the only columns decrypted, not the entire flash cache cacheline. If the first predicate does not return any matches, then the columns for the second predicate are not accessed. This feature has up to 30% faster encrypted smart scans and reduced storage server CPU usage.
For example, consider the following query:
SELECT ename FROM emp
WHERE job LIKE ‘%VP’ AND sal + bonus > 500000
Only the projected columns (ename) and one or more predicated columns are decrypted. The salary in data regions that do not have a VP are not accessed.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Smart In-Memory Columnar Cache with Row IDs
Oracle Exadata System Software evaluates the predicate in a DML statement to determine the rows of interest. These rows are sent back from the storage servers to the database servers, which then uses the ROWIDs to update the rows. This feature expands this functionality to return ROWIDs for data stored in columnar compression format.
For example, consider the following SQL query:
UPDATE weather_history SET temp = (temp – 32) * (5/9)
WHERE country = 'ENGLAND'
Smart scan is used to find the rows that satisfy country = 'ENGLAND' using ROWID values returned from a previous phase of the command processing.
This feature can improve the performance of DML statements by up to 5x.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Database release 18.3.0.0.180717DBRU
Smart In-memory Columnar Cache with Chained Rows
Wide tables, large rows or row migration on update can create chained rows. Smart In-memory Columnar Cache works by creating optimized Single Instruction Multiple Data (SIMD) representation for chained rows when the head and tail pieces fit in the same 1 MB region. This results in up to 3x faster scans for wide tables.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Fast Smart In-memory Columnar Cache Creation
The run-time analyzer finds the best compression algorithm, reusing the dictionary created in-line during the analysis. This provides faster symbol lookup and insert using the new dictionary data structure. The Single Instruction Multiple Data (SIMD) optimized columnar cache creation reuses data from the first pass scan in the flash cache. The result is much faster reads from the flash cache, with an up to 35% improved speed in columnar cache creation.
This feature applies to Exadata Hybrid Columnar Compression tables only.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Update a Single SNMP User Definition
This feature adds new syntax to ALTER CELL and ALTER DBSERVER commands to allow SNMP V3 users to be added, altered, and deleted individually instead of configuring all users in one command string. This features makes it easier to change the password for an individual SNMP user.
ALTER CELL {
| SHUTDOWN SERVICES { RS | MS | CELLSRV | ALL } [IGNORE REDUNDANCY]
| RESTART SERVICES { RS | MS | CELLSRV | ALL } [IGNORE REDUNDANCY]
| RESTART BMC
| STARTUP SERVICES { RS | MS | CELLSRV | ALL }
| LED {ON | OFF}
| DONOTSERVICELED {ON | OFF [FORCE]}
| VALIDATE { MAIL | SNMP | CONFIGURATION }
| VALIDATE SYSLOGCONF selector.node
| CONFIGUREBMC
| BBU { DROP FOR REPLACEMENT | REENABLE }
| { snmpuser=((user_clause)[,(user_clause)[,..]]) | snmpuser.name=(user_clause) }
| attribute_name = attribute_value
[, attribute_name = attribute_value]...
}
ALTER DBSERVER {
| SHUTDOWN SERVICES { RS | MS | ALL }
| RESTART SERVICES { RS | MS | ALL }
| RESTART BMC
| STARTUP SERVICES { RS | MS | ALL }
| LED { ON | OFF }
| VALIDATE { MAIL | SNMP | CONFIGURATION }
| VALIDATE SYSLOGCONF facility.priority
| CONFIGUREBMC
| { snmpuser=((user_clause)[,(user_clause)[,..]]) | snmpuser.name=(user_clause) }
| attribute_name = attribute_value
[, attribute_name = attribute_value]...
}
Securing Storage Server Software Processes with Memory Protection Keys
Memory Protection Keys is a hardware feature found in Oracle Exadata Database Machine X7-2 and newer systems. Memory Protection Keys provide a thread local permission control on memory pages without incurring the high cost of Page Table Entry (PTE) modifications and Translation Look-aside Buffer (TLB) flushes.
The Exadata storage server process that performs block IO (cellsrv) and the processes that perform smart scans (celloflsrv) are now enhanced to run with memory protection keys. This feature is enabled out of the box with no tuning needed. Each thread in these processes needs to obtain access to the appropriate memory protection key before it can access the data. Any access to a piece of memory that does not have the correct key traps the process. This enhances the security and robustness of the storage server processes by eliminating a class of potential memory corruptions.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Exadata Database Machine X7-2
Default XFS File System for X8M Servers
To further improve the security of Oracle Exadata Database Machine, the new X8M servers will use the XFS file system for bare metal, KVM guests, and storage servers.
New partitions are created for /home, /tmp, /var, and /var/log/audit.
Minimum requirements:
Oracle Exadata System Software release 19.3.0
Oracle Exadata Database Machine X8M servers
Oracle Linux Kernel to Unbreakable Enterprise Kernel 5 and Oracle Linux Distribution Upgraded to Oracle Linux 7.7
This release upgrades Oracle Linux to Unbreakable Enterprise Kernel (UEK) 5, (4.14.35-x.el7uek.x86_64). Oracle Linux Kernel UEK5 introduces support for persistent memory, ROCE, and KVM on Exadata. The Linux distribution is upgraded to Oracle Linux 7.7.
Minimum requirements to upgrade to Oracle Exadata System Software release 19.3.0:
Oracle Grid Infrastructure:
19.4.0.0.0.190716 with patches for 30177140 and 29391849
18.7.0.0.0.190716 with patches for 29637526 and 29391849
12.2.0.1.0.190716 with the patch for 29637526
12.1.0.2.0.190716 with the patch for 30139389
Oracle Database:
19.4.0.0.0.190716 with the patches for 30177140 and 30165493
18.7.0.0.0.190716 with the patch for 30034456
12.2.0.1.0.181016
12.1.0.2.0.180831
11.2.0.4.0.180717
Software Certification Ends for Exadata Database Machine X2 Servers
Software certification ends for Oracle Exadata Database Machine X2 servers. This includes the following servers:
Oracle Exadata Database Machine X2-8 Database Server
Sun Fire X4170 Oracle Database Servers
Sun Fire X4170 M2 Oracle Database Servers
Sun Fire X4800 Oracle Database Servers
Oracle Exadata Storage Server X2-2
Oracle Exadata Storage Server with Sun Fire X4270 M2 Servers
Oracle Exadata Storage Server with Sun Fire X4275 Servers
