CSC/ECE 506 Spring 2011/ch6a ep

INTRODUCTION TO MEMORY HIERARCHY ORGANIZATION
Write-Miss Policies and Prefetching

Write miss policies and prefetching are two strategies that are used by multiprocessors to achieve optimal performance for memory accesses. Write miss policies can help eliminate write misses and therefore reduce bus traffic. This reduces the wait-time of all processors sharing the bus. Prefetching assures that a CPU has data blocks in cache to be read when processing large data files and streaming data. When CPUs or cores share a cache, prefetched data in a shared cache is available to all processes on those cores processing the data.

This article begins by highlighting the variety of multicore processors on the market today that have hierarchal memory structures and shared caches. It then explores the write miss policies and prefetching techniques that these multiprocessors can use to take advantage of these architectures.

Recent Architectures and their Cache Characteristics

Multicore processor architectures with both shared and private caches are common in desktop and server platforms from a variety of manufacturers. Below is a table that illustrates the variety of core and cache designs in multiprocessors from four major manufacturers:

Table 1: Recent Architectures and their Cache Characteristics [1][2][3][4]
Company	Processor	Cores	L1 Cache	L2 Cache	L3 Cache
AMD	Athlon 64 X2	2	128 KB x 2	512 KB x 2 1MB x 2
AMD	Athlon 64 FX	2	128 KB x 2	1MB x 2
AMD	Athlon X2	2	128 KB x 2	512 KB x 2
AMD	Phenom X3	3	128 KB x 3	512 KB x 3	2 MB
AMD	Phenom X4	4	128 KB x 4	512 KB x 4	2 MB
AMD	Athlon II X2	2	128 KB x 2	512 KB x 2 1MB x 2
AMD	Athlon II x3	3	128 KB x 3	512 KB x 3
AMD	Athlon II X4	4	128 KB x 4	512 KB x 4
AMD	Phenom II X2	2	128 KB x 2	512 KB x 2	6 MB
AMD	Phenom II X3	3	128 KB x 3	512 KB x 3	6 MB
AMD	Phenom II X4	4	128 KB x 4	512 KB x 4	4-6 MB
AMD	Phenom X6	6	128 KB x 4	512 KB x 4	6 MB
Intel	Pentium D	2	12+16KB x 2	2 MB
Intel	Celeron E	2	32 KB x2	512 -1 MB
Intel	Pentium D	2	32 KB x2	1 MB
Intel	Core 2 Duo	2	32 KB x2	2 - 4 MB
Intel	Core 2 Quad	4	32 KB x2	8 MB
Intel	Atom 330	2	32+24KB x 2	512 MB x 2
Intel	Pentium E	2	32 KB x 2	2 MB
Intel	Core 2 Duo E	2	32 KB x 2	3-6 MB
Intel	Core 2 Quad	4	32 KB x 4	2-6 MB x 2
Intel	Core i3	2	32+32 KB x 2	256 KB x2	4 MB
Intel	Core i5 - 6 Series	2	32+32 KB x 2	256 KB x2	4 MB
Intel	Core i5 - 7 Series	4	32+32 KB x 4	256 KB x 4	8 MB
Intel	Core i5 - 2400 Series Core i5 - 2500 Series	4	32+32 KB x 4	256 KB x 4	6 MB
Intel	Core i7 - 8 Series	4	32+32 KB x 4	256 KB x 4	8 MB
Intel	Core i7 - 9 Series	4	32+32 KB x 4	256 KB x 4	8 MB
Intel	Core i7 - 970	6	32+32 KB x 6	256 KB x 6	12 MB
Sun	UltraSPARC T1	8	8 K x 8 Inst. 16 K x 8 Data	3 MB
Sun	SPARC64 VI	2	128 K x 2 Inst. 128 K x 2 Data	5 MB
Sun	UltraSPARC T2	8	8 K x 4 Inst. 16 K x 4 Data	4 MB
Sun	SPARC64 VII	4	64 K x 4 Inst. 64 K x 4 Data	6 MB
Sun	SPARC T3	8	8 K x 8 Inst. 16 K x 8 Data	6 MB
Sun	SPARC64 VII+	2	128 K x 2 Inst. 128 K x 2 Data	12 MB
IBM	Power5	2	64 K x 2 Inst. 64 K x 2 Data	4 MB x 2	32 MB
IBM	Power7	4, 6, or 8	32+32 KB x C	256 kB x C	4 - 32 MB x C

Note: The 'x #' in the L1 and L2 columns indicates that this cache is for each core.

Cache Write Policies[5]

In section 6.2.3, cache write hit policies and write miss policies were explored. The write hit policies, write-through and write-back, determine when the data written in the local cache is propagated to lower-level memory. The write miss policies, write-allocate and no-write-allocate, determine if a memory block is stored in a cache line after the write occurs. Note that since a miss occurred on write, the block is not already in a cache line.

Write-miss policies can also be expressed in terms of write-allocate, fetch-on-write, and write-before hit. These can be used in combination with the same write-through and write-back hit policies to define the complete cache write policy.

Fetch-on-Write

On a write miss, the memory block is not found in the cache, and the memory block containing the address to be written is fetched from the lower level memory hierarchy before the write proceeds. Note that this is different from write-allocate. Fetch-on-write determines if the block is fetched from memory, and write-allocate determines if the memory block is stored in a cache line. Certain write policies may allocate a line in cache for data that is written by the CPU without retrieving it from memory first.

No-Fetch-on-Write

On a write miss, the memory block is not fetched first from the lower level memory hierarchy. Therefore, the write can proceed with out having to wait for the memory block to be returned.

Write-Before-Hit

On a write, the write proceeds before the cache determines if a hit or miss occurred. In this scenario, the tag and the data can be written simultaneously, but it incurs an immediate bus transaction for each write by the processor.

No-Write-Before-Hit

On a write, the write waits until the cache determines if the block being written to is in the cache or not. This may avoid a bus transaction by allowing the processor to write to the cache multiple times before the cache line is flushed to memory.

Write-Miss Policy Combinations

In practice, these policies are used in combination to provide an over-all write policy. Four combinations of these three write miss policies are relevant, and are expressed in the table below:

The two combinations of fetch-on-write and no-write-allocate are blocked out in the above diagram. They are considered unproductive since the data is fetched from lower level memory, but is not stored in the cache. Temporal and space locality suggest that this data will be used again, and will therefore need to be retrieved a second time.

Three combinations named write-validate, write-around, and write-invalidate result in 'eliminated misses' which in general will yield better cache performance if the overhead to manage the policy remains low. They all use no-fetch-on-write.

Write-Validate

The combination of no-fetch-on-write and write-allocation is referred to as 'write-validate'. It writes the data into the cache line without fetching the corresponding block from memory first. The assumption is that the block will be written to memory at a later time. It requires additional overhead, or dirty bits, to track what bytes have been changed in the cache, and lower level memories must be able can process only the changed portions of the lines.

Write-Around

The combination of no-fetch-on-write, no-write-allocate, and no-write-before-hit is referred to as a 'write-around' since it bypasses the higher level cache, writes directly to the lower level memory, and leaves the existing block line in the cache unaltered on a hit or a miss. This strategy shows performance improvements when the data that is written will not be reread in the near future. Since we are writing before a hit is detected, the cache is written around for both hits and misses.

Write-Invalidate

The combination of write-before-hit, no-fetch-on-write, and no-write-allocate is referred to as 'write-invalidate' because the line is invalidated on the miss. This is because the copy that exists in lower level memory after the write miss differs from the one in the cache. For write hits, though, the data is simply written into the cache using the cache hit policy. Thus, for hits, the cache is not written around.

Fetch-on-write

When used with write-allocate, fetch-on-write stores the block and processes the write to cache as expected.

Prefetching in Contemporary Parallel Processors

Prefetching is a technique that can be used in addition to write-hit and write-miss policies to improved the utilization of the cache. Microprocessors, such as those listed in Table 1, support prefetching logic directly in the chip.

Intel Core i7 [6]

The Intel 64 Architecture, including the Intel Core i7, includes both instruction and data prefetching directly on the chip.

The Data Cache Unit prefetcher is a streaming prefetcher for L1 caches that detects ascending access to data that has been loaded very recently. The processor assumes that this ascending access will continue, and prefetches the next line.

The data prefetch logic (DPL) maintains two arrays to track the recent accesses to memory: one for the upstreams that has 12 entries, and one for downstreams that has 4 entires. As pages are accessed, their address is tracked in these arrays. When the DPL detects an access to a page that is sequential to an existing entry, it assumes this stream will continue, and prefetches the next cache line from memory.

AMD [7]

The AMD Family 10h processors use a stream-detection strategy similar to the Intel process described above to trigger prefetching of the next sequential memory location into the L1 cache. (Previous AMD processors fetched into the L2 cache, which introduced the access latency of the L2 cache and hindered performance.)

The AMD Family 10h can prefetch more than just the next sequential block when a stream is detected, though. When this 'unit-stride' prefetcher detects misses to sequential blocks, it can trigger a preset number of prefetch requests from memory. For example, if the preset value is two, the next two blocks of memory will be prefetched when a sequential access is detected. Subsequent detection of misses of sequential blocks may only prefetch a single block. This maintains the 'unit-stride' of two blocks in the cache ahead of the next anticipated read.

AMD contends that this is beneficial when processing large data sets which often process sequential data and process all the data in the stream. In these cases, a larger unit-stride populates the cache with blocks that will ultimately be processed by the CPU. AMD suggests that the optimal number of blocks to prefetch is between 4 and 8, and that fetching too many or fetching too soon has impaired performance in empirical testing.

The AMD Family 10h also includes 'Adaptive Prefetching', a hardware optimization that triggers prefetching when the demand stream catches up to the prefetch stream. In this case, the unit-stride is increased to assure that the prefetcher is fetching at a rate sufficient enough to continuously provide data to the processor.

References

[1]: http://www.techarp.com/showarticle.aspx?artno=337
[2]: http://en.wikipedia.org/wiki/SPARC
[3]: http://en.wikipedia.org/wiki/POWER7
[4]: http://en.wikipedia.org/wiki/POWER5
[5]: “Cache write policies and performance,” Norman Jouppi, Proc. 20th International Symposium on Computer Architecture (ACM Computer Architecture News 21:2), May 1993, pp. 191–201. http://www.hpl.hp.com/techreports/Compaq-DEC/WRL-91-12.pdf
[6]: "Intel® 64 and IA-32 Architectures Optimization Reference Manual", Intel Corporation, 1997-2011. http://www.intel.com/Assets/PDF/manual/248966.pdf
[7]: "Software Optimization Guide for AMD Family 10h and 12h Processors", Advanced Micro Devices, Inc., 2006–2010. http://support.amd.com/us/Processor_TechDocs/40546.pdf