I have found that enabling Jumbo frames resulted in decreased read and write performance with the following hardare:
Server:
IBM x3550 M3 (7944AC1) with 1 6c Intel Xeon cpu X5650 and 12gb of memory
dual port QLogic iSCSI hba QLA406x with bios 1.14
Switch:
Two HP 2810-24G
Storage:
IBM DS3300 dual controller with 512MB cache
Configuration:
Jumbo frames are enabled at the QLogic Card in the FAST!util setup
Jumbo frames are enable for the VLANS on the HP 2810-24 switches
Jumbo frames are enabled on the four host ports of the DS3300.
I tested with a 2 disk SAS RAID 1 array on the DS3300.
The test Virtual machine was windows XP with the C: drive on local SAS disks and it's E: drive an 8GB buslogic scsi drive on the iSCSI storage array added via VMware.
I use microsoft's SQLIO as a benchmarking tool and record the mb/s for each test.
Without JUMBO FRAMES
8KB Random Write | 64kb Random Write | 128kb Random write | 256kb Random write | 8KB SEQUENTIAL Write | 64KB SEQUENTIAL Write | 128KB SEQUENTIALWrite | 256KB SEQUENTIAL Write | 8KB RANDOM Read | 64KB RANDOM Read | 128KB RANDOM Read | 256KB RANDOM Read | 8KB SEQUENTIAL Read | 64KB SEQUENTIAL Read | 128KB SEQUENTIAL Read | 256KB SEQUENTIAL Read |
<!--
BODY,DIV,TABLE,THEAD,TBODY,TFOOT,TR,TH,TD,P { font-family:"Liberation Sans"; font-size:x-small }
-->
7.16 | 39.73 | 56.82 | 77.78 | 17.07 | 78.98 | 84.13 | 88.49 | 68.43 | 111.49 | 111.67 | 111.45 | 67.94 | 111.72 | 111.76 | 111.8 |
WITH JUMBO FRAMES:
<!--
BODY,DIV,TABLE,THEAD,TBODY,TFOOT,TR,TH,TD,P { font-family:"Liberation Sans"; font-size:x-small }
-->
<!--
BODY,DIV,TABLE,THEAD,TBODY,TFOOT,TR,TH,TD,P { font-family:"Liberation Sans"; font-size:x-small }
-->
8KB Random Write | 64kb Random Write | 128kb Random write | 256kb Random write | 8KB SEQUENTIAL Write | 64KB SEQUENTIAL Write | 128KB SEQUENTIALWrite | 256KB SEQUENTIAL Write | 8KB RANDOM Read | 64KB RANDOM Read | 128KB RANDOM Read | 256KB RANDOM Read | 8KB SEQUENTIAL Read | 64KB SEQUENTIAL Read | 128KB SEQUENTIAL Read | 256KB SEQUENTIAL Read |
<!--
BODY,DIV,TABLE,THEAD,TBODY,TFOOT,TR,TH,TD,P { font-family:"Liberation Sans"; font-size:x-small }
-->
6.73 | 38.38 | 56.81 | 68.83 | 17.66 | 73.41 | 76.11 | 76.31 | 66.66 | 91.35 | 92.89 | 90.14 | 66.27 | 91.33 | 92.89 | 90.14 |
Here is the test program:
<!--
BODY,DIV,TABLE,THEAD,TBODY,TFOOT,TR,TH,TD,P { font-family:"Liberation Sans"; font-size:x-small }
-->
Download SQLIO from Microsoft: |
|
|
|
run the installer |
|
it will put the program in C:\Program Files (x86)\SQLIO\ |
|
|
create a bat or cmd file called SQLIO.cmd ( any name will do) and paste this into it: |
|
sqlio -kW -s10 -frandom -o8 -b8 -LS -Fparam.txt timeout /T 60 > results.txt |
sqlio -kW -s360 -frandom -o8 -b64 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kW -s360 -frandom -o8 -b128 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kW -s360 -frandom -o8 -b256 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kW -s360 -fsequential -o8 -b8 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kW -s360 -fsequential -o8 -b64 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kW -s360 -fsequential -o8 -b128 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kW -s360 -fsequential -o8 -b256 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -frandom -o8 -b8 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -frandom -o8 -b64 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -frandom -o8 -b128 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -frandom -o8 -b256 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -fsequential -o8 -b8 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -fsequential -o8 -b64 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -fsequential -o8 -b128 -LS -Fparam.txt timeout /T 60 >> results.txt |
sqlio -kR -s360 -fsequential -o8 -b256 -LS -Fparam.txt >> results.txt |
I'm assuming that I must not have something configured properly, Does someone have similiar hardware have better results?
