0% found this document useful (0 votes)
30 views5 pages

AEPWatch Overview

Download as pdf or txt
Download as pdf or txt
Download as pdf or txt
You are on page 1/ 5

AEPWatch Overview

Contents
Introduction .................................................................................................................................................. 3
Tool Usage..................................................................................................................................................... 3
Explanation of metrics .................................................................................................................................. 4
Introduction
AEPWatch is a tool that monitors and reports the behavior of the AEP DIMM. It consists of 2 tools; the
data collector is called AEPWatch and a second visualization tool named generate_plot.sh uses gnuplot
to visualize the collected data.

Tool Usage
AEPWatch "<sampling interval>" "<number of samples>" -f outputfile.csv

Example: "AEPWatch 1 100" reads the counters every 1 second 100 times.

generate_plot.sh outputfile.csv: See below for an example of the charts generated.


Explanation of metrics
bytes_read (derived) :
bytes_written (derived):

Number of bytes transacted by the read and write operations.


Note: The total number of bytes transacted in any sample is computed as bytes_read (derived) + 2 *
bytes_written (derived).

Formula:
bytes_read : (read_64B_ops_received - write_64B_ops_received) * 64
bytes_written: write_64B_ops_received * 64

read_hit_ratio (derived) :
write_hit_ratio (derived):
Measures the efficiency of the prefetcher. Range of 0.0-1.0.

Formula:
read_hit_ratio : (ddrt_read_ops - sxp_read_ops) / ddrt_read_ops
write_hit_ratio: (ddrt_write_ops - sxp_write_ops) / ddrt_write_ops

wdb_merge_percentage (derived):
Measures efficiency of the write merges. Range of 0.0-1.0.

0.75 : indicates 100% sequential write traffic


> 0.75 : indicates writing to 64B addresses that are still in the WDB (never had to go to media)
1 or ~1: likely writing to a specific address or small range of addresses (fitting in WDB) for long periods of
time.
Current implementation will force the WDB entries to be flushed to media after 3ms in the WDB. Once
the WDB is scheduled to go to media it is invalidated from the WDB.

Formula:
CONSTANT = 10us (clock rate * number of clocks for DDP)
sxp_write_ops_adj = sxp_write_ops - (sample_time / CONSTANT)
wdb_merge_percentage: (ddrt_write_ops - sxp_write_ops_adj) / ddrt_write_ops
sxp_read_ops (derived) :
sxp_write_ops (derived):

Number of operations performed by the microcontroller to the physical media.

Formula:
sxp_read_ops : (read_64B_ops_received - write_64B_ops_received) / 4
sxp_write_ops: write_64B_ops_received / 4

read_64B_ops_received :
write_64B_ops_received:

Number of read and write operations generated by the microcontroller. Each operation transacts a
64byte operation. These operations includes commands transacted on behalf of the microcontroller for
maintenance as well as the commands transacted by the CPU.

ddrt_read_ops :
ddrt_write_ops:

Number of read and write operations received from the CPU (memory controller), for the 2LM and App
Direct Mode partitions.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy