Electrical computers and digital processing systems: memory – Storage accessing and control – Specific memory composition
Reexamination Certificate
1999-12-09
2003-09-02
Yoo, Do Hyun (Department: 2187)
Electrical computers and digital processing systems: memory
Storage accessing and control
Specific memory composition
C711S147000, C711S154000, C710S035000, C710S060000
Reexamination Certificate
active
06615308
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention relates generally to data transfer, and in particular to data transfer through a bus or a system controller.
2. Discussion of the Related Art
As workstation and server systems use more and more DRAM memory, the physical and logical lengths of the channels between the system memory controller and the DRAM devices themselves are becoming longer. As a result, the transit time of data between the memory controller and the DRAMs is increasing. When this happens on a system that uses a single channel for both read and write data, “turning the bus around” between the read data and the write data results in more substantial periods during which no data is valid at one end of the chain or the other. This is a waste of DRAM channel bandwidth.
The traditional technique for minimizing wasted bandwidth is to enforce long write bursts and long read bursts. This increases the amount of time that data is being pipelined through the channel and decreases the amount of time that is spent in transition.
Unfortunately, a memory controller that solves the problem of wasted bandwidth in this way is quickly saddled with the problem of elevated system memory read latency, because the performance of a device demanding data is often degraded more by latency increase than by memory bandwidth deficiencies.
So far, this problem has been solved with static methods that pre-specify the minimum size of the write burst. However, optimizing the system for low latency (worst case of 2 additional cycles) can cause system bandwidth to be degraded to 40% of the theoretical bandwidth; conversely, optimizing for only a 28% bandwidth loss can cause a latency hit of up to 8 cycles. Both of these circumstances are aggravated in the worst case: when the memory controller is receiving requests that cause it to perform the same number of DRAM write transactions as DRAM read transactions.
SUMMARY OF THE INVENTION
In one embodiment, monitoring data traffic through a memory controller; and dynamically and automatically selecting a burst length for data transactions through a memory controller in response to the monitored data traffic is disclosed.
REFERENCES:
patent: 5224213 (1993-06-01), Dieffenderfer et al.
patent: 5359593 (1994-10-01), Derby et al.
patent: 5918072 (1999-06-01), Bhattacharya
patent: 5948081 (1999-09-01), Foster
patent: 6014384 (2000-01-01), Weberhofer
patent: 6128233 (2000-10-01), Yu et al.
patent: 6256684 (2001-07-01), Klein
patent: 6282208 (2001-08-01), Bowcutt et al.
patent: 6393500 (2002-05-01), Thekkath
patent: 6397287 (2002-05-01), Brown et al.
patent: 6411946 (2002-06-01), Chaudhuri
patent: EP 1043902 (2000-11-01), None
Blakely , Sokoloff, Taylor & Zafman LLP
Chace Christian P.
Intel Corporation
Yoo Do Hyun
LandOfFree
Method and apparatus for regulating write burst lengths does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for regulating write burst lengths, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for regulating write burst lengths will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3093847