Selective Memory

Scheme would make new high-capacity data caches 33 to 50 percent more efficient.


Conventional computers have microprocessors mounted on a package, with embedded electrical leads. The package snaps into the motherboard and information go between the processor and the PC’s fundamental memory bank through the leads.

As transistors counts have gone up in chips, it affects the processor’s performance. Thus,  the moderately moderate association between the processor and fundamental memory has turned into the main obstacle to enhancing PCs’ execution. That’s the reason, chip manufacturers started using DRAM right on the chip package.

DRAM is fundamentally different from the high-capacity type of cache. In addition, most cache-management schemes don’t use it efficiently. To address this, researchers from MIT, Intel, and ETH Zurich developed a new cache-management scheme called Banshee.

The processing unit in a chip usually has a table that maps the virtual addresses used by each program to the actual addresses of data stored in main memory. This system works by adding three bits of data to each entry in the table. First bit shows whether the information at that virtual address can be found in the Measure reserve. Second and third bits demonstrate its area with respect to some other information things with a similar hash file.

Scientists noted the bandwidth in this in-package DRAM can be five times higher than off-package DRAM. It also promises to improve the data rate of in-package DRAM caches by 33 to 50 percent.

Xiangyao Yu, a postdoc in MIT’s Computer Science and Artificial Intelligence Laboratory said, “In the entry, you need to have the physical address, you need to have the virtual address, and you have some other data. That’s already almost 100 bits. So three extra bits is a pretty small overhead.”

If one of a chip’s cores pulls a data item into the DRAM cache, the other cores won’t know about it. Sending messages to all of a chip’s cores every time any one of them updates the cache consumes a good deal of time and bandwidth. So Banshee introduces another small circuit, called a tag buffer, where any given core can record the new location of a data item it caches.

The tag buffer work by checking the request sent via main memory by any core. It then checks whether the requested tag is one whose location has been remapped. When the Buffer is filled completely, the Banshee notifies the chips’ cores that they need to update their virtual-memory tables. Then it clears the buffer and starts over.

Scientists noted, “The buffer is small, only 5 kilobytes, so its addition would not use up too much valuable on-chip real estate. The time required for one additional address lookup per memory access is trivial compared to the bandwidth savings Banshee affords.”


See stories of the future in your inbox each morning.