Warning: session_start(): Cannot send session cookie - headers already sent by (output started at /home/httpd/html/www.ccii.co.za/company/press_releases/ppb_cci1.html:2) in /home/httpd/html/www.ccii.co.za/toptemplate.php on line 4

Warning: session_start(): Cannot send session cache limiter - headers already sent (output started at /home/httpd/html/www.ccii.co.za/company/press_releases/ppb_cci1.html:2) in /home/httpd/html/www.ccii.co.za/toptemplate.php on line 4
PacedPacketBatch Technology Improves Network Performance [CI Press Release, 1999-08-11]

Our Products
   Board-Level Products
   System-Level Products
   Remote Monitoring
   and Control
Product Photo Gallery
Product Support
   User Manuals and Software Drivers
   RMA and Software Support
   Standard Terms & Conditions
   Software Library
About CI Systems
   Company Profile (pdf)
   Real-Time House (pdf)
   Production Facility (pdf)
   Company Capabilities
   Company Services
   CI Systems Management
   Press Releases
R & D
Press Releases

Return Material Authorisation  (RMA)

Search CI Systems using Google :


   Products       Presentations       Search    Where We Are Contact Us    Home   
PacedPacketBatch Technology Improves Network Performance
Publication: CI Press Release Issued: Date: 1999-08-11 Reporter: CI Systems

CI Press Release


One of the major causes of diminished network performance is the interrupt process triggered by incoming data packets.

However SysKonnect, Europe's leader in FDDI network cards (IDC 1997) currently represented in South Africa by high-speed, mission-critical networking specialist, CI Systems, has developed PacedPacketBatch technology to minimise these so-called interrupt handling problems.

PacedPacketBatch technology allows the recently launched SK-NET GE Gigabit Ethernet server adapters to dynamically adapt to interrupt requests as well as optimise interrupt behaviour to suit existing load conditions.

According to CI Systems managing director, Richard Young, on conventional NICs, every incoming data packet triggers an interrupt of the host CPU. This is done so the CPU can be alerted to activity requests from devices in the bus.

However, it also means that any currently running process will be interrupted while the CPU responds to the interrupt request. This interrupt process can be quite complex, as many events usually result in more work for the CPU, more time required to handle the request and less performance for the whole system.

"When handling an interrupt request, the host must first localise the source of an interrupt before the CPU can execute the appropriate interrupt service routine," says Young.

"However, before the interrupt service routine can be processed, the then current register values and other states must be saved to memory for later use. Once the interrupt service routine is processed, the register values and states are retrieved from memory, reset, and the original interrupted process is then resumed.

"The relatively simple process of handling an interrupt usually has a very large impact on system performance. The by-product of this interrupt handling work is typically seen as high CPU utilisation and low system throughput. Even more significant, these conditions get worse as a system gets busier, resulting in significant overhead, or load, for the host system.

"PacedPacketBatch is a sophisticated form of dynamic interrupt moderation that is able to reduce load on the CPU and immediately improve overall system performance."

Young says PacedPacketBatch technology groups individual interrupt sources together with an interrupt controller located onboard the adapter. This grouping capability reduces the number of times running processes are actually interrupted for service by the CPU, and the load on the overall system is therefore reduced.

Even more effective, the SK-NET GE interrupt controller is also dynamic. SysKonnect's PacedPacketBatch technology adjusts the interrupt grouping effect in step with changing traffic conditions to minimise interrupt loading conditions, he says.

When network traffic is low, a smaller number of individual interrupts are grouped by this "auto-adjust" mechanism in PacedPacketBatch. When traffic rates are high, larger numbers of interrupts can be group together for greater efficiency. The overall effect of this dynamic interrupt moderation is seen through reduced CPU utilisation and lower transfer latencies at higher data rates. This is particularly valuable in high-performance servers.

"SK-NET GE adapters also minimise host CPU utilisation by performing CPU-intensive tasks such as calculating TCP, UDP and IP checksums independent of the host operating system," says Young.

These services are provided by the SysKonnect ASIC on the SK-NET GE adapter via hardware thereby offloading the host CPU, he says, while conversion of little endian/big endian addressing by the SysKonnect ASIC also serves to reduce CPU utilisation and free up host resources for more productive system tasks.

"The ASIC of the SK-NET GE Gigabit Ethernet server adapter can independently adapt the word order of data to be passed to and from system memory without assistance from the host CPU - one more reason why the SK-NET GE adapter is particularly well suited for use in systems where every spare CPU cycle is at a premium," concludes Young.

Please refer to www.syskonnect.com or www.ccii.co.za for more information on the SK-NET GE Gigabit Ethernet and PacedPacketBatch.