\n
Memory and TLB Parameters:
\nΔt_mem
): 100 ns
Δt_TLB, L1-D = 1 ns\nΔt_TLB, L2 = 5 ns\n
\nρ_TLB, L1-D = 30%\nρ_TLB, L2 = 98%\n
\nBehavior:
\nΔt_TLB, L1-D
.Δt_TLB, L2
.I’ve tried breaking this into three cases:
\nCase 1: L1-D TLB Hit (30%
):
Time = Δt_TLB, L1-D = 1 ns\n
\nCase 2: L1-D Miss, L2 TLB Hit (70% × 98%
):
Time = Δt_TLB, L1-D + Δt_TLB, L2 = 1 ns + 5 ns = 6 ns\n
\nCase 3: Both L1-D and L2 TLB Miss (70% × 2%
):
Time = Δt_TLB, L1-D + Δt_TLB, L2 + 4 × Δt_mem\nTime = 1 ns + 5 ns + 4 × 100 ns = 406 ns\n
\nFinally, I believe the overall EMAT is the weighted sum of these cases based on their probabilities, but I’m unsure if I’m combining them correctly.
\nI’d appreciate any clarification or suggestions to ensure my understanding is correct.
\nThanks!
\n","author":{"@type":"Person","name":"Bishop_1"},"upvoteCount":1,"answerCount":1,"acceptedAnswer":null}}Reputation: 71
I’m working on a problem involving the calculation of effective memory-access time (EMAT) for a system with multi-level TLBs, and I want to confirm if my understanding is correct. Here’s the problem setup:
Memory and TLB Parameters:
Δt_mem
): 100 ns
Δt_TLB, L1-D = 1 ns
Δt_TLB, L2 = 5 ns
ρ_TLB, L1-D = 30%
ρ_TLB, L2 = 98%
Behavior:
Δt_TLB, L1-D
.Δt_TLB, L2
.I’ve tried breaking this into three cases:
Case 1: L1-D TLB Hit (30%
):
Time = Δt_TLB, L1-D = 1 ns
Case 2: L1-D Miss, L2 TLB Hit (70% × 98%
):
Time = Δt_TLB, L1-D + Δt_TLB, L2 = 1 ns + 5 ns = 6 ns
Case 3: Both L1-D and L2 TLB Miss (70% × 2%
):
Time = Δt_TLB, L1-D + Δt_TLB, L2 + 4 × Δt_mem
Time = 1 ns + 5 ns + 4 × 100 ns = 406 ns
Finally, I believe the overall EMAT is the weighted sum of these cases based on their probabilities, but I’m unsure if I’m combining them correctly.
I’d appreciate any clarification or suggestions to ensure my understanding is correct.
Thanks!
Upvotes: 1
Views: 92
Reputation: 1
Your approach is fine.
Instead of three seperate cases you could also use this, which is not better but just another way of doing it:
EMAT = (L1 hit rate × L1 time) +
(L1 miss rate × L2 hit rate × L2 time) +
(L1 miss rate × L2 miss rate × full miss penalty)
Upvotes: 0