1. Field of the Invention
The invention relates to caching where portions of data are stored in slower main memory and are transferred to faster memory between one or more processors and the main memory. The system is such that an individual cache system must announce to other cache systems or check with other cache systems which may possibly contain a copy of a given cached location prior to or upon modification or appropriation of data at a given cached location, and includes provisions for determining when the data stored in a particular memory location may be replaced.
2. Background Art
In multi-processor systems with secondary shared cache, e.g., Level 2 (L2) cache, the shared cache holds lines beyond the actual residency of the cache in subordinate caches, e.g., a Level 1 (L1) cache. This is the value of a secondary cache; it contains a superset of all subordinate caches. This reduces traffic from main memory.
When a processor, e.g., processor A, requests a line for store (“EX” or exclusive request) that line is installed in both the L1 and the L2 caches as exclusive to processor A. When another processor, e.g., processor B, requests the line, processing is elongated if the line is held exclusive to processor A.
However, as is often the case, that processor A is no longer using the line, and, in fact the line has aged out of L1 and is only in L2. Nevertheless, the line is still marked exclusive to processor A. Thus extra processing, such as cross investigating and cross invalidating, is required to remove this line from the L1 cache.
In many cases as described here, this line no longer exists in L1. This unnecessarily elongates the fetch request for processor B, reducing performance. Thus a need exists to speed up this process.
The method and system described herein provides caching where portions of data are stored in slower main memory and are transferred to faster memory between one or more processors and the main memory. As a general rule, the cache is such that an individual cache system must communicate to other associated cache systems, or check with such cache systems, to determine if they contain a copy of a given cached location prior to or upon modification or appropriation of data at a given cached location. The cache further includes provisions for determining when the data stored in a particular memory location may be replaced.
By way of background, a line must exist in L1 cache to be used by any processor. Now, when a line exists in both L1 and L2, the state of the line is said to match in both caches. That is, the line is either marked exclusive, shared, modified the same in both levels. If a line ages out of L1 and only resides in L2, then that line is no longer in use by the original processor. More recently used lines have replaced that line. However, the L2 cache continues to hold the line in a state of “Exclusive” to that processor.
The method and system described herein changes the state of the line in the L2 cache when the line is to be replaced in the L1 cache. The state changes from exclusive to a particular identified processor to exclusive to no processor. Thus, when an exclusive line ages out of a particular processor, the line state in L2 will change from exclusive to the particular identified processor to “none”, and no cross interrogation or invalidation is necessary for any L1 cache, thereby improving overall system performance.
Caching where portions of data are stored in slower main memory and are transferred to faster memory between one or more processors and the main memory. The cache is such that an individual cache system must communicate to other associated cache systems, or check with such cache systems, to determine if they contain a copy of a given cached location prior to or upon modification or appropriation of data at a given cached location. The cache further includes provisions for determining when the data stored in a particular memory location may be replaced.
As described herein, when an exclusive line ages out of an L1 cache, the L1 sends a signal to the L2 cache, indicating that the line no longer exists in the L1 cache. This causes the L2 cache to be updated to indicate that the line is “disowned.” That is, the ownership is changed from the particular processor to “unowned.”
By way of background, despite improvements in technology, microprocessors are still much faster than main memory. Memory access time is increasingly the bottleneck in overall application performance. As a result, an application might spend a considerable amount of time waiting for data. This not only negatively impacts the overall performance, but the application cannot benefit much from a processor clock-speed upgrade either.
One way to overcome this problem is to insert a small high-speed buffer memory between the processor and main memory. Such a buffer is generally referred to as cache memory, or cache for short. Such a buffer or cache is illustrated in
The system 101 first copies the data needed by the processor 111 from main memory 131 into the L2 cache 121, and then from the L2 cache 121 to the L1 cache 113 and into a register (not shown) in the processor 111. Storage of results is in the opposite direction. First the system copies the data from the processor 111 into the L1 cache 113, and from the L2 cache 121. Depending on the cache architecture details, the data is then immediately copied back to memory 131 (write-through), or deferred (write-back). If an application needs the same data again, data access time is reduced significantly if the data is still in the L1 cache 113 and L2 cache 121 or only the L2 cache 121. To further reduce the cost of memory transfer, more than one element is loaded into cache. The unit of transfer is called a cache block or cache line. Access to a single data element brings an entire line into the cache. The line is guaranteed to contain the element requested.
Latency and bandwidth are two metrics associated with caches and memory. Neither of them is uniform, but is specific to a particular component of the memory hierarchy. The latency is often expressed in processor cycles or in nanoseconds, while bandwidth is usually given in megabytes per second or gigabytes per second.
In practice the latency of a memory component is measured as the time it takes to fetch the first portion of a unit of transfer (typically a cache line). As the speed of a component depends on its relative location in the hierarchy, the latency is not uniform. As a rule of thumb, it is safe to say that latency increases when moving from L1 cache 113 to L2 cache 121 to main memory 131.
Some of the memory components, the L1 cache 113 for example, may be physically located on the processor 111. The advantage is that their speed will scale with the processor clock. It is, therefore, meaningful to express the latency of such components in processor clock cycles, instead of nanoseconds.
On some microprocessors, the integrated (on-chip) caches, as L1 cache 113, do not always run at the speed of the processor. They operate at a clock rate that is an integer quotient (½, ⅓, and so forth) of the processor clock.
Cache components external to the processor do not usually, or only partially, benefit from a processor clock upgrade. Their latencies are often given in nanoseconds. Main memory latency is almost always expressed in nanoseconds.
Bandwidth is a measure of the asymptotic speed of a memory component. This number reflects how fast large bulks of data can be moved in and out. Just as with latency, the bandwidth is not uniform. Typically, bandwidth decreases the further one moves away from the processor 111.
If the number of steps in a data fetch can be reduced, latency is reduced.
When an exclusive line ages out of an L1 cache 313, the L1 cache 313 sends a signal to the L2 cache 321, indicating that the line no longer exists in the L1 cache 313. This causes the L2 cache 321 to be updated to indicate that the line is “disowned.” That is, the ownership is changed from the particular processor to “unowned.”
Consider the cache shown in
The cache index selects the tag to be tested to be see if the desired block is in the cache. The next step is index selection. In the case of direct mapping, the data is read and sent to the processor 311 in parallel with the tag being read and checked.
After reading the tag from the cache 321, it is compared to the tag portion of the block address from the processor 311. To be sure that the tag and the data line both contain valid information, the validity data (as exclusive to a processor on “unowned”) must be read and verified. If the cache data is valid, the processor 311 is signaled to load the data from the L2 cache 321.
In the case of a write, the data is also written to a write buffer, for writing to main memory.
In the case of a cache miss, the L1 cache 313 sends a stall signal to the processor 311, and a read to L2 cache 321 and then to the main memory 331. Reading from main memory replaces a line in cache, e.g., typically the least recently used line, including the data, the tags, and the validity information.
According to the method and system of the invention, when an exclusive line ages out of an L1 cache, and the L1 cache sends a signal to the L2 cache, this indicates that the line no longer exists in the L1 cache, thereby causing the L2 cache to be updated to indicate that the line is “disowned.” That is, the ownership is changed from the particular processor to “unowned.”
Looking at
Additionally, eliminating cross interrogate processing avoids sending a cross invalidate to an L1 cache 213a or 213b for a line that is to be replaced in the L2 cache 221. Ordinarily, when a line ages out of L2 cache 221, that line must also be invalidated in the L1 cache 213a or 213b. This maintains a subset rule between L1 213a or 213b and L2 221 caches. These two invalidates disrupt normal processing at the L1 cache 213a or 213b. If the line no longer exists in the L1 cache 213a or 213b, this disruption is unnecessary and negatively impacts performance.
While the method and system have been described with respect to exclusive lines, it is to be understood that the method and system described herein also avoids cross interrogates and cross invalidates for shared lines when one updates an L2 cache directory 221 for shared lines that age out of the L1 cache 221a or 221b.
If an L1 shared cache implementation maintains a separate bit for each processor to identify which L1 caches “hold the line”, each time a line ages out of the L1 cache 211a or 211b, the respective “CP holds the line” bit in the L2 cache 221 could be reset. Cross interrogates and cross invalidates need to be performed only to CP's having their “CP holds line” bit active. If all bits are inactive, all cross interrogates and cross invalidates are avoided. In this way, L2 cache 221 age out replacements avoid unnecessary cross invalidates to L1 caches 213a or 213b. This too improves performance by avoiding unnecessary disruptions at L1 caches 213a or 213b.
A further aspect of the invention is a program product comprising computer readable program code to configure and control a computer system. The computer system comprises a plurality of microprocessors, where each microprocessor has an associated L1 cache 213a or 213b, an L2 cache 221 associated to the L1 caches 213a and 213b, and main memory 231. The program product causes the caches to store portions of data for transfer to and from the processors and the main memory, and to further cause an individual cache communicating to other associated caches to check with the said other caches, to determine if they contain a copy of a given cached location prior to reading data from, reading data to, or modifying cached data. The computer code also determines when the data stored in a particular cache memory location may be replaced.
Specifically, the computer code writes changes to the state of the line in the L2 cache when the line is to be replaced in the L1 cache, and also changes the state of the line from exclusive to a particular identified processor to exclusive to no processor.
While the invention has been described with certain preferred embodiments and exemplifications, it is not intended to limit the scope of the invention thereby, but solely by the claims appended hereto.