The TLB misses have very enormous impact on the overall performance for the processor. In order to achieve higher performance, recent TLB designs tend to provide more entries, larger page size, or even superpage mechanism. However, very few attempts have been made for the context switching issue which causes TLB cold-start misses very seriously. In fact, in our knowledge, almost no research really seriously considers this issue. In this thesis, we presented two novel TLB mechanisms to reduce the miss rate in context switching. The one is a TLB structure suitable for large page size, such as 1MB size. The other is a novel TLB structure which reduces the miss rate in context switching with small page sizes, such as 4KB, 8KB or 16KB sizes.
Our studies of new novel TLB structure focus on how to reduce TLB size for small page size TLB. We combine both the features of Lee’s dual TLB and our original TLB with many TLB banks in our new structure. We use one shared conventional small page (4KB size) TLB and 16 large page (16KB size) promotion-TLB banks. Each bank has translations for each process and only the shared TLB need to be flushed when context switching. The shared TLB works together with only one of the promotion-TLB banks simultaneously. In this structure we can reduce the miss rate in context switches with small page size by keeping the translation in each bank and with the help of the shared TLB which is sufficient to reduce the miss rate. We also proposed some mechanisms to implement the new TLB structure and how to modify OS to support it. In addition, we improve the utilization of TLB entries by making use of Lee et al.’s promotion mechanism and sending back the least recently used entry from the one current TLB bank into the shared TLB.
In order to make reasonable and fairly comparisons of our design with others and to exclude the influence of multiple parameters varying simultaneously in OS, we just consider both the best case situation and worst case situation of our design. It is shown that the relative performance improvement of miss rate of the new design is almost better or equal to that of other new proposed TLB structure or conventional fully-associative TLB structure. Our proposed TLB can even achieve about 1.3 of the relative improvement of miss rate in average with 4KB page size. In fact, if the page size is simply increase to 16 KB, the proposed TLB can achieve almost to 1.45.
That means our TLB can obtain much more improvement with larger page size than other structures. Furthermore, the cost of the new design is not much higher than other new structures or conventional fully-associative TLB. Thus the suggested new TLB structure is worthy to be implemented for contemporary or future high performance processors.
In order to achieve high performance, recent TLB research tends to support large page sizes or multiple page sizes. For example, Naohiko Shimizu and Ken Takatori proposed a Linux superpage kernel for Alpha, Sparc64 and IA32 [20]. In conventional approaches, to support multiple page sizes needs lots of OS modifications. In our research, we take the advantage of complete-subblock TLB structure without OS help to support two different page sizes. However, the complete-subblock TLB needs much more hardware cost than the superpage TLB with OS help. We have already begun to find a solution to integrate the superpage mechanism with low hardware cost. Furthermore, we would integrate our proposed structure with prefetching mechanism to reduce the miss rate. We believe that still lots of works should be done in this field.
Appendix A:
Detailed Relative Improvement Figures
This appendix includes relative improvement of miss rate for twelve SPEC2000 benchmarks. The behavior of individual workloads is shown here. We can look at the comparison of simulation results for data TLB (DTLB) and instruction TLB (ITLB) with different TLB structures in more detail. Usually, the ITLB has better relative improvement of miss rate than the DTLB because the ITLB exhibits greater locality than the DTLB. These figures below from Figure 19 to Figure 30 correspond to the Figure 17, where we average the relative improvement of miss rate for DTLB and ITLB.
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 19: The relative improvement of miss rate in gzip.
DTLB
1 0.998041572 1.002862382
1.082478457
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.006994888
1.213290113
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.998041572 1.002862382
1.034987307
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.011786276 1.011786276
0.994 1.0081.01 1.012 1.014
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1.000279966 1.033983794
1.19819863
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.006994888
1.213290113
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1.000279966
1.033983794
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1.000273299 1.040807621
1.380480905
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.017731806
2.651588065
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1.000273299
1.040807621
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.017731806 1.017731806
0.99
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 20: The relative improvement of miss rate in vpr.
DTLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.321318229 1.321525336
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.028472054 1.028721683
0.9850.99 0.9951 1.0051.01 1.0151.02 1.0251.03 1.035
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1.24344613 1.415998583
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.028472054 1.028721683
0.9850.99 0.9951 1.0051.01 1.0151.02 1.0251.03 1.035
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1.24344613 1.243397767
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.045507135 1.045809625
0.97
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.087511124
2.028776978
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.349109349 1.349109349
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.087511124 1.087511124
0.94
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 21: The relative improvement of miss rate in gcc.
DTLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.999332068
1.004736154
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.999332068 1.004736154
0.944230386
0.91 0.920.93 0.94 0.950.96 0.970.98 0.991 1.01
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.963151291 0.957602219
0.945726884
0.910.92 0.93 0.940.95 0.960.97 0.980.991 1.01
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.025673804
1.200947986
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.963151291 0.957602219
0.912660545
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.999596835
1.024958661
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.039614474
1.399142574
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.999596835
1.024958661
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 22: The relative improvement of miss rate in crafty.
DTLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.015754393
1.380455668
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.015754393
1.154499541
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.010041791
0.73776792
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.056057866
1.936339523
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.010041791
0.768764302
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.056057866 1.060275962
0.960.97 0.980.991 1.011.02 1.031.04 1.051.06 1.07
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.064417178
2.20177665
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.03267254 1.03839055
0.98
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB 1.021.04 1.06 1.081.1 1.121.14
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 23: The relative improvement of miss rate in vortex.
DTLB
1
0.980392157 0.988142292
0.939849624
0.910.9 0.920.93 0.940.95 0.960.97 0.980.991 1.01
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.980392157 0.988142292
0.939849624
0.910.9 0.920.93 0.940.95 0.960.97 0.980.991 1.01
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 24: The relative improvement of miss rate in lucas.
DTLB
1
0.957856713 0.959589867
1.090472927
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1.001474201
1.167239404 1.331156107
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.957856713 0.959589867
1.090472927
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1.001474201
1.167239404
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.003192338
1.267137097
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.008438819
1.125588697
0.920.94 0.960.981 1.021.04 1.061.081.1 1.121.14
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1.008438819 1.008438819
0.994
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.003264418
1.11622276
0.940.96 0.981 1.02 1.041.06 1.081.1 1.12 1.14
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.28496732 1.439238653
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1.28496732 1.28496732
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 25: The relative improvement of miss rate in twolf.
DTLB
1 1 1.13091075
1.352609565
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.077962417 1.217505701
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.13091075 1.124754272
0.9
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.077962417 1.082332762
0.94
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.165786948
1.553388747
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.094266939
1.33582296
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.165786948 1.168871783
0.9
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.920.94 0.960.981 1.021.04 1.061.081.1 1.121.14
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.227224763
1.803100775
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.429720776 1.429720776
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.227224763 1.227224763
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 26: The relative improvement of miss rate in swim.
DTLB
1 0.999970849 1.000206536
1.001535261
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.001736111
2.028119508
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.999944705
1.002779986 1.002731796
0.99850.999 0.9995 1 1.00051.001 1.0015 1.002 1.00251.003
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
0.9850.99 0.995 1 1.0051.01 1.015 1.02 1.0251.03
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 0.999944705
1.002779986
1.00075262
0.99850.999 0.9995 1 1.00051.001 1.0015 1.002 1.00251.003
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 0.999762443
1.004585995 1.005276797
0.997
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.006525285
1.045762712
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.999762443
1.004585995
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
0.9850.99 0.9951 1.0051.01 1.0151.02 1.0251.03 1.035
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 0.999970849
1.000206536
Conventional TLB
Complete-subblock TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 27: The relative improvement of miss rate in perlbmk.
DTLB
1 1 1.012891344
1.072124756
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.960.97 0.980.991 1.011.02 1.031.04 1.051.06 1.07
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.9850.980.99 0.9951 1.0051.01 1.0151.02 1.0251.03 1.035
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.025806452
1.127659574
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB 1.011.02 1.031.04 1.051.06 1.07
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.10982659 1.254901961
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.065217391
1.232704403
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.10982659 1.103448276
0.94
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 28: The relative improvement of miss rate in applu.
DTLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.000420765
1.249326765
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.000420765
1.043848096
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.000740796
1.160068699
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB 0.99981 1.0002 1.0004 1.0006 1.00081.001 1.0012
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
0.891323058 0.917948548
0.863996973
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.9994999
1.043219543
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
0.820.84 0.86 0.880.9 0.92 0.940.96 0.981 1.02
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.9994999
1.043219543 1.043219543
0.97
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 29: The relative improvement of miss rate in equake.
DTLB
1 0.964788275 0.993589989
0.762445479
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.307726966 1.307726966
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.998485855 1.005116959
1.041112883
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.998485855
1.005116959
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1.456628478 1.456866918
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.24077283 1.308092802
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1.24077283 1.24077283
0
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Best situation
Page Size is 4k
(PS = 4K) Worst situation
Best situation
Page Size is 8k
(PS = 8K) Worst situation
Best situation
Page Size is 16k
(PS = 16K) Worst situation
Figure 30: The relative improvement of miss rate in mgrid.
DTLB
1 0.99804664
0.986360097 0.985585477
0.975
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.000198098
4.552750225
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.99804664
0.986360097
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.000198098
1.014364641
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
0.991880497 0.999250609
1.030718176
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1 1.000555324
1.031634698
0.9850.980.99 0.9951 1.0051.01 1.0151.02 1.0251.03 1.035
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 1 1.000555324
1.010515984 1.0081.01 1.012
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel
1 0.995592258 1.00053664
1.068623113
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1.000427259 1.001497006
1.084278768
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 0.995592258 1.00053664
1.03882641
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
1 1.000427259 1.001497006
1.032862814
Conventional TLB Complete-subblock
TLB Lee et al.'s TLB Our new novel TLB
Reference
[1] Advanced Micro Devices, Inc., Software Optimization Guide for AMD AthlonTM 64 and AMD OpteronTM Processors, November 2004.
[2] D. Burger and T.M. Austin, “The SimpleScalar Tool Set Version 2.0,” Technical Report 1342, Computer Sciences Department, University of Wisconsin, Madison, WI, 1997.
[3] R. Case and A. Padegs. Architecture of the IBM System/370, ch 51, pp.830-855, McGraw-Hill Book Company, New York, 1982.
[4] David Channon and David Koch, “Performance Analysis of Re-configurable Partitioned TLBs,” in Proceedings of the 30th Hawaii Int’l Conf. on System Sciences, Vol. 5, pp.168-177, 1995.
[5] J. B. Chen, A. Borg, and N. P. Jouppi. “A simulation Based Study of TLB Performance.” Pages 114-123, 1992.
[6] Chang-Jiu Chen, Wei-Min Cheng, Chi-Wen Chang, Wen-Chiuan Liao, “A Novel TLB Architecture To Reduce Miss Rate In Context Switching,” in the Proceedings of
[6] Chang-Jiu Chen, Wei-Min Cheng, Chi-Wen Chang, Wen-Chiuan Liao, “A Novel TLB Architecture To Reduce Miss Rate In Context Switching,” in the Proceedings of