Note: Results recorded in this page are with a modified version of Postmark (v1.5).

Machine Characteristics:
cpu0: Intel Pentium 4 (686-class), 2391.26 MHz
I-cache 12K uOp cache 8-way, D-cache 8 KB 64B/line 4-way
L2 cache 512 KB 64B/line 8-way
ITLB 4K/4M: 128 entries
DTLB 4K/4M: 64 entries
total memory = 382 MB
avail memory = 365 MB
root file system type: ffs
At the beginning of the test, FileSystem occupancy was:
Filesystem Size Used Avail Capacity Mounted on
/dev/wd0a 19G 13G 4.9G 72% /
kernfs 1.0K 1.0K 0B 100% /kern


Benchmarking Results:

.....LEGEND.....
TT = Total Time
TPS = Transactions per second
FCPS = Files Created per second
RR = Read Rate


Case A:
#######
Kernel: Original NetBSD kernel (v3.0)
Comments:In this case, Postmark was excuted, with no background "dd" process running.

PARAMETERS: disable_dd=[1], program_repeat_count=[5] file_size_low=[500], file_size_high=[10000], simultaneous=[20000], transactions=[25000],
subdirectories=[10], read_block_size=[512], write_block_size=[512]

Result:TT->[14.650604]secs,TPS->[20551.267191],FCPS->[1365.131431],RR->[8884477.936882]
Result:TT->[8.440659]secs,TPS->[38722.228418],FCPS->[2369.483236],RR->[15420948.530204]
Result:TT->[8.287755]secs,TPS->[38747.795250],FCPS->[2413.198749],RR->[15705455.578742]
Result:TT->[8.239690]secs,TPS->[39216.116883],FCPS->[2427.275783],RR->[15797071.006312]
Result:TT->[8.339383]secs,TPS->[38735.427732],FCPS->[2398.258960],RR->[15608225.212825]

Each of the above results are for single execution of the process. Since "program_repeat_count=[5]", we have five results.

Case B:
#######
Kernel: Original NetBSD kernel (v3.0)
Comments: In this case, Postmark was excuted, with background "dd" process running.

PARAMETERS: disable_dd=[0], program_repeat_count=[5] file_size_low=[500], file_size_high=[10000], simultaneous=[20000], transactions=[25000],
subdirectories=[10], read_block_size=[512], write_block_size=[512]

1000+0 records in
1000+0 records out
1048576000 bytes transferred in 39.029 secs (26866586 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 42.015 secs (24957181 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 41.690 secs (25151739 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 41.925 secs (25010757 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 42.432 secs (24711915 bytes/sec)

Result:TT->[16.637432]secs,TPS->[13166.716437],FCPS->[1202.108595],RR->[7823501.126857]
Result:TT->[9.002543]secs,TPS->[24396.765574],FCPS->[2221.594498],RR->[14458466.679915]
Result:TT->[8.630853]secs,TPS->[22498.139404],FCPS->[2317.268061],RR->[15081124.426520]
Result:TT->[8.526174]secs,TPS->[25848.957305],FCPS->[2345.718021],RR->[15266280.983710]
Result:TT->[8.628388]secs,TPS->[24466.603575],FCPS->[2317.930070],RR->[15085432.875758]

Observe the decrease in TPS and in the RR values. Penalty suffered (in TPS) due to write floods is approximately 42%.

Case C:
#######
Kernel: Original NetBSD kernel (v3.0) compiled with UVM_CCA.
Comments: In this case, Postmark was excuted, with background "dd" process running.

PARAMETERS: disable_dd=[0], program_repeat_count=[5] file_size_low=[500], file_size_high=[10000], simultaneous=[20000], transactions=[25000],
subdirectories=[10], read_block_size=[512], write_block_size=[512]

1000+0 records in
1000+0 records out
1048576000 bytes transferred in 74.031 secs (14164012 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 75.412 secs (13904630 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 69.857 secs (15010321 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 71.930 secs (14577728 bytes/sec)
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 74.050 secs (14160378 bytes/sec)

Observe throttling of write requests.

Result:TT->[20.127512]secs,TPS->[11453.751356],FCPS->[993.664791],RR->[6466917.918121]
Result:TT->[9.524172]secs,TPS->[21098.435171],FCPS->[2099.920077],RR->[13666591.489528]
Result:TT->[33.376658]secs,TPS->[20939.215969],FCPS->[599.221168],RR->[3899820.287579]
Result:TT->[33.170998]secs,TPS->[26224.627020],FCPS->[602.936336],RR->[3923999.151307]
Result:TT->[33.282655]secs,TPS->[28136.237270],FCPS->[600.913599],RR->[3910834.877806]

This result is somewhat unclear on how UVM_CCA is affecting the net throughput. The execution time has increased but only in two
cases we see around 16% performance improvement. The RR has decresed dramatically. This and other cases are currently
under active investigation.


###################