摘要 |
Method for control of traffic generated by a processor where a pre-fetch request from a processor 102, 104, or 106 for data or instructions from a memory hierarchy 108 includes a pre-fetch identifier in the request. This identifier flags the request as a pre-fetch request rather than a time-critical request. Based on the identifier, the memory hierarchy issues an abort response via generating logic 128 to the processor at times of high traffic (monitored by assessment logic 130) and pre-fetch traffic is suppressed. On receipt of an abort response, the device deletes records of the pre-fetch request. If the data / instruction is later required, a new request is issued at a higher priority than the original request. Prefetch identifier is selected according to speculative or definite request. Assessment of traffic levels is dynamically updated and includes measurements of: Quality of Service (QoS), requests accepted in a window, bandwidth consumed. |