Re: [Iometer-devel] Checking in AIO fix for Linux dynamo

SourceForge Headquarters 1320 Columbia Street Suite 310 San Diego, CA 92101 +1 (858) 422-6466

Hi Marty,

Thanks for updating test results.

I attached a new patch for 2006 version (rev2006) and v1.1.0-rc1. 
This patch will fix the problem in prepare disk and CPU util update.

I usually tested linux dynamo on Linux dynamo VM running VMWare ESX.
I compared Iometer reported numbers and IOPS numbers from esxtop (ESX performance monitoring tool).
I didn't find such performance difference in my tests (local disk, FC SAN -- random/sequential mix)

But I suggest you turn off linux I/O scheduler (from CFQ to noop).
With CFQ I/O scheduler, I also observed a big difference between Iometer reported numbers and ESXTOP reported numbers.

Here is an instruction to turn off CFQ scheduler in Linux.

-----------------------------------------------------------------
Here is instructions to turn off CFQ I/O scheduler.
To set the scheduler in Grub.
If we take noop as the target default scheduler for the system, the /boot/grub/menu.lst kernel entry would look like this:
title CentOS (2.6.18-128.4.1.el5)
root (hd0,0)
kernel /vmlinuz-2.6.18-128.4.1.el5 ro root=/dev/VolGroup00/LogVol00 elevator=noop
initrd /initrd-2.6.18-128.4.1.el5.img
Having the elevator entry in place, the system will set the I/O scheduler to the specified one on every boot.
In short, add "elevator=noop" in kernel boot parameter line in /boot/grub/menu.lst.
-----------------------------------------------------------------

Hope this helps.

Thanks.
-JK

-----Original Message-----
From: Marty Schlining [mailto:msc...@dd...] 
Sent: Thursday, September 29, 2011 10:08 AM
To: Jinpyo Kim
Subject: RE: [Iometer-devel] Checking in AIO fix for Linux dynamo

Hi Jinpyo,

I am giving your changes a try. 

I compiled the IOMeter GUI (Win32 Release) using Microsoft VC++ 2008 on Windows 7. I compiled dynamo on a linux x86_84 platform (Centos 5.5). No problems, there.

I was also able to run everything the way I expected.

My concern is that the results posted in the IOMeter GUI do not match what is being measured at the target. I'm not sure how to begin debugging this.

Perhaps I should try a later version of IOMeter with your changes?

For a 512k 100% sequential read to my target (12 LUNs), IOMeter reports a total of 1.4 GB/s (1423 MB/s), while the target is showing 2.0 GB/s (which is accurate based on previous measurements with xdd). IO sizes looks correct at the target.

Target IO measurements per LUN:

Virtual Disk Counters:  Elapsed time = 1475.216 seconds
  Idx  IOs/sec  KiB/sec   KiB/IO Fwd IO/s Fwd KiB/s| IOs/sec  KiB/sec   KiB/IO Fwd IO/s Fwd KiB/s|
--------------------------------------------------------------------------------------------------
    0        0        0        0        0        0 |     323   165712      512        0        0 |
    1      338   173412      512        0        0 |       0        0        0        0        0 |
    2        0        0        0        0        0 |     366   187689      512        0        0 |
    3      339   173869      512        0        0 |       0        0        0        0        0 |
    4        0        0        0        0        0 |     325   166456      512        0        0 |
    5      340   174273      512        0        0 |       0        0        0        0        0 |
    6        0        0        0        0        0 |     326   167141      512        0        0 |
    7      337   172666      512        0        0 |       0        0        0        0        0 |
    8        0        0        0        0        0 |     325   166473      512        0        0 |
    9      338   173213      512        0        0 |       0        0        0        0        0 |
   10        0        0        0        0        0 |     325   166590      512        0        0 |
   11      346   177442      512        0        0 |       0        0        0        0        0 |

 Total 2.064 GB/s

That's pretty awesome for dynamo on Linux. Can't wait to see what it will do on a larger system.

Best Regards,
Marty Schlining
DataDirect Networks

-----Original Message-----
From: Jinpyo Kim [mailto:jk...@vm...]
Sent: Friday, September 23, 2011 2:27 PM
To: Daniel Scheibli
Cc: Iom...@li...
Subject: Re: [Iometer-devel] Checking in AIO fix for Linux dynamo

Hi,

I quickly created  and attached a patch for 1.1 RC build. 
It compiles OK, but not tested with 1.1 RC Iometer controller.
I will do it by next week.

Thanks.
-JK

-----Original Message-----
From: Daniel Scheibli [mailto:da...@sc...]
Sent: Friday, September 23, 2011 1:39 AM
To: Jinpyo Kim
Cc: Vedran Degoricija; Iom...@li...
Subject: Re: [Iometer-devel] Checking in AIO fix for Linux dynamo

Hi Jinpyo,

thanks for the patch! Aravind contacted me some months ago about it, so its awesome to see it happen. If you can provide the path against the recent RC that would be of great help so checking and integrating becomes easier.

Thanks,
Daniel

PS: The mail with the attachment got hold up by the mailing list,
    but I approved it, so it should show up soon.

> Yes, there is little difference in Linux code in recent 1.1 RC build 
> compared to 2006 version.
> If you need a patch for recent RC build, I will send it too.
>
> Thanks.
> -JK
>
> From: Vedran Degoricija [mailto:ve...@ya...]
> Sent: Thursday, September 22, 2011 3:30 PM
> To: Jinpyo Kim; Iom...@li...
> Subject: Re: [Iometer-devel] Checking in AIO fix for Linux dynamo
>
> Hi Jinpyo,
>
> This is good news! I was not aware of your efforts. We have been 
> needing this for a long time.
>
> I suspect that today's Linux code has not changed much since 2006, but 
> we'd need to check that is the case, or else you need to make a patch 
> based on the latest code in SVN.
>
> Also, would you pass your code changes to us in a tarball for review?
>
> Thanks,
> Ved
>
>
> From: Jinpyo Kim <jk...@vm...>
> To: "Iom...@li..."
> <Iom...@li...>
> Sent: Thursday, September 22, 2011 3:07 PM
> Subject: [Iometer-devel] Checking in AIO fix for Linux dynamo Hi,
>
> A few months back, Aravind B. (from VMware) contacted whether we can 
> contribute AIO fix for linux dynamo.
> Iometer tool was widely used for many I/O tests at VMware or by our 
> customers.
>
> But existing released version and recent RC build had the same 
> problems not  issuing multiple outstanding I/Os properly.
> We have a fixed linux dynamo version (used for a while), but it would 
> be better integrated in next release of Iometer.
>
> I already created a patch from 2006 released version src tree.
> Please let me how to check it in.
>
> Thanks.
> -JK
>
>
> ----------------------------------------------------------------------
> -------- All the data continuously generated in your IT infrastructure 
> contains a definitive record of customers, application performance, 
> security threats, fraudulent activity and more. Splunk takes this data 
> and makes sense of it. Business sense. IT sense. Common sense.
> http://p.sf.net/sfu/splunk-d2dcopy1
> _______________________________________________
> Iometer-devel mailing list
> Iom...@li...<mailto:Iom...@li...
> orge.net> https://lists.sourceforge.net/lists/listinfo/iometer-devel
>
> ----------------------------------------------------------------------
> -------- All the data continuously generated in your IT infrastructure 
> contains a definitive record of customers, application performance, 
> security threats, fraudulent activity and more. Splunk takes this data 
> and makes sense of it. Business sense. IT sense. Common sense.
> http://p.sf.net/sfu/splunk-d2dcopy1___________________________________
> ____________
> Iometer-devel mailing list
> Iom...@li...
> https://lists.sourceforge.net/lists/listinfo/iometer-devel
>