dslreports logo
 
    All Forums Hot Topics Gallery
spc
Search similar:


uniqs
638

PToN
Premium Member
join:2001-10-04
Houston, TX

PToN

Premium Member

Measuring IOs

Hello,

We are about to install a few database servers in our virtual environment and i am looking for ways to get some IO data to see what kind of performance we can see.

We are having the following DBs: MongoDB, MSSQL, MySQL, and Rocket Unidata (formerly IBM UniData).

What do you use to get this info? I know i can use dd, iotop in linux; and recently came across SQLIO. Could SQLIO be use for disk performance or does it do some SQL specific read/write when it runs?

Thanks.

hairspring
join:2007-11-23
Oakville, ON

hairspring

Member

SQLIO is suitable for base disk performance measurements. If you do want to test some SQL specific workloads for OLTP or Data Warehousing you can use SQLIOSIM.

Now I kinda wonder why it's called SQLIO at all. It really has nothing to do with SQL..

quaker
Premium Member
join:2001-12-27
Rocky River, OH

quaker to PToN

Premium Member

to PToN
IOMeter is another tool to produce loads »www.iometer.org/

PToN
Premium Member
join:2001-10-04
Houston, TX

PToN

Premium Member

Anyone used IOzone?

I am getting way different results from iozone and SQLIO...

SQLIO: 800MB file - 64K - 1 thread
READ Random - 148.17 IOs/sec - 9.26 MBs/sec
WRITE Random - 529.86 IOs/sec - 33.11MBs/sec
 

And
IOzone: 800MB file - 64K - 1 thread
READ Random - 448.47 MBs/sec
Write Random - 297.81 MBs/sec
 

Both were done to the same disk...
amungus
Premium Member
join:2004-11-26
America

amungus

Premium Member

Haven't heard of / tried IOzone, but that is interesting.
What kind of storage is this against - "local" to host, SAN, RDM? Across what kind of connection - ("local" / direct) SAS, iSCSI, FibreChannel?

PToN
Premium Member
join:2001-10-04
Houston, TX

PToN

Premium Member

Well i first ran them in a Windows 2008R2 VM. The disk was presented to the VM with XenSever and the array is FC with 15k disks.

But then i decided to run it on my local PC, SSD and i get the same discrepancy with all of the same params

SQLIO:
READ: 3486.08 IOs/sec - 217.88 MBs/sec
WRITE: 3746.32 IOs/sec - 234.14 MBs/sec
 

IOzone:
READ: 3106.31 MBs/sec
WRITE: 2458.08 MBs/sec
 

So idk. i thought it would be a much closer difference between the 2 programs. But that's way too much..

dennismurphy
Put me on hold? I'll put YOU on hold
Premium Member
join:2002-11-19
Parsippany, NJ

dennismurphy to PToN

Premium Member

to PToN
iometer is the gold standard, IMO.

IOZone benchmarks file systems; iometer does disk devices. The iometer data is MUCH more useful to me ...

quaker
Premium Member
join:2001-12-27
Rocky River, OH

quaker to PToN

Premium Member

to PToN
also make when testing, your using a test file that is bigger than the cache on SAN. otherwise it will skew your results.

PToN
Premium Member
join:2001-10-04
Houston, TX

PToN

Premium Member

using fio seems more in line with SQLIO... And dd supports SQLIO and fio results. Maybe i configured the IOzone test wrong.

Thanks for ya'lls input

exocet_cm
Writing
Premium Member
join:2003-03-23
Brooklyn, NY

exocet_cm to PToN

Premium Member

to PToN
SolarWinds Virtual Machine Monitors. Provides us with live IOP data on our VMs and datastores.

You could also use the advanced performance data on the VMs in vCenter too.