IBM 990 - Manual (Page 232)

220 IBM
^
zSeries 990 Technical Guide
The z990 servers have performance improvements on all workload environments, from
traditional to e-business on demand. Comparing to the z900 turbo servers, the z990 has
improved all major components:
Maximum number of assigned processors, from 16 to 32
Processor cycle time, from 1.09 ns to 0.83 ns
L2 caches, from 32 MB per 20 PUs to 32 MB per 12 PUs
Maximum memory size, from 64 GB to 256 GB
STI bandwidth, from 1 GBps to 2 GBps per STI
Maximum number of STIs, from 24 to 48 STIs
Maximum I/O bandwidth, from 24 GBps to 96 GBps
Maximum number of channels, from 256 to 1024
Additional performance improvements for e-business
z990 has also further performance improvements for e-business application environments:
zSeries Application Assist Processors (zAAPs), which are designed to operate
asynchronously with the CPs to execute Java programming under control of IBM Java
Virtual Machine (JVM) for logical partitions running z/OS. The IBM JVM processing cycles
can be executed on the configured zAAPs with no anticipated modifications to the Java
applications.
IEEE Floating Point: Used by Java and C/C++ applications, the new Binary Floating Point
unit halves the number of cycles required on previous servers.
Secondary level Translation Lookaside Buffer (TLB): A secondary cache for Dynamic
Address Translation, for both the instruction and data caches, increases the number of
buffer entries by a factor of eight.
CP Assist for Cryptographic Function (CPACF): Implemented on each PU, the assist
function uses five new instructions for symmetrical clear key cryptographic encryption and
encryption operations, to accelerate the encryption and decryption of SSL transactions,
and VPN encrypted data transfers.
In addition, the following improvements for specific areas are also implemented on z990:
Compression Unit:
The Compression Unit is integrated with the CP Assist for Cryptographic Function,
benefiting from combining the use of buffers and interfaces. It is implemented on each PU
and provides excellent hardware compression performance.
Checksum offload for IPV4 packets when in QDIO mode for Linux and z/OS:
Checksum Offload provides the capability of calculating the Transmission Control Protocol
(TCP), User Datagram Protocol (UDP), and Internet Protocol (IP) header checksums.
Checksum verifies the correctness of files. By moving the checksum calculations to a
Gigabit or 1000BASE-T Ethernet feature, host CPU cycles are reduced and performance
is improved. It is supported by the OSA-Express GbE and 1000BASE-T Ethernet features
when operating at 1 Gbps.
Multi-book structure
The multiple book structure introduced with the z990 servers offers more flexibility, capacity,
and scalability to the system.