Hardware Recommendations?

[Robotbeat]

rogerk, what sort of performance have you been getting with larger test file sizes? Like, twice or thrice as big as your ram?

[rogerk]

1 thread writing for 20 secs to file d:\testfile.dat
using 4KB random IOs
enabling multiple I/Os per thread with 64 outstanding
buffering set to use hardware disk cache (but not file cache)
size of file d:\testfile.dat needs to be: 41943040000 bytes
current file size: 4194304000 bytes
need to expand by: 37748736000 bytes
expanding d:\testfile.dat ... done.
using specified size: 40000 MB for file: d:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 147.68
MBs/sec: 0.57
latency metrics:
Min_Latency(ms): 5
Avg_Latency(ms): 369
Max_Latency(ms): 7260
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100

this was a 40GB File from our exchange 2003 on citrix xen.
what a performance do you aspect?

regards
roger

[1parkplace]

Ouch, those results seem fairly bad to me... Especially the latency.

[rogerk]

i made another test with a non production vitual machine
testfile 40GB

C:\Program Files (x86)\SQLIO>sqlio.exe -kW -s20 -frandom -o64 -b4 -LS -Fparam.tx
t
sqlio v1.5.SG
using system counter for latency timings, 3579545 counts per second
parameter file used: param.txt
file g:\testfile.dat with 1 thread (0) using mask 0x0 (0)
1 thread writing for 20 secs to file g:\testfile.dat
using 4KB random IOs
enabling multiple I/Os per thread with 64 outstanding
size of file g:\testfile.dat needs to be: 41943040000 bytes
current file size: 10485760 bytes
need to expand by: 41932554240 bytes
expanding g:\testfile.dat ... done.
using specified size: 40000 MB for file: g:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 33119.70
MBs/sec: 129.37
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 1
Max_Latency(ms): 403
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 84 15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1

so i hope this help

i think the test above was made on the wrong machine.
our exchange has a 300GB database...

[Robotbeat]

i made another test with a non production vitual machine
testfile 40GB

C:\Program Files (x86)\SQLIO>sqlio.exe -kW -s20 -frandom -o64 -b4 -LS -Fparam.tx
t
using specified size: 40000 MB for file: g:\testfile.dat
IOs/sec: 33119.70
MBs/sec: 129.37
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 1
Max_Latency(ms): 403

That's more like it!

We need to get a 10GbE card to test with...

[rogerk]

C:\Program Files (x86)\SQLIO>sqlio.exe -kW -s20 -frandom -o64 -b4 -BH -LS -Fpara
m.txt
sqlio v1.5.SG
using system counter for latency timings, 3579545 counts per second
parameter file used: param.txt
file g:\testfile.dat with 1 thread (0) using mask 0x0 (0)
1 thread writing for 20 secs to file g:\testfile.dat
using 4KB random IOs
enabling multiple I/Os per thread with 64 outstanding
buffering set to use hardware disk cache (but not file cache)
size of file g:\testfile.dat needs to be: 83886080000 bytes
current file size: 41943040000 bytes
need to expand by: 41943040000 bytes
expanding g:\testfile.dat ... done.
using specified size: 80000 MB for file: g:\testfile.dat
initialization done
CUMULATIVE DATA:
throughput metrics:
IOs/sec: 30826.99
MBs/sec: 120.41
latency metrics:
Min_Latency(ms): 0
Avg_Latency(ms): 1
Max_Latency(ms): 787
histogram:
ms: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24+
%: 78 21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1

this was with a 80GB file

in a non virtual system this should be 30% better.

regards
roger

[Robotbeat]

How are reads?