Project Home
Project Home		 Documents
Documents		 Wiki
Wiki		 Discussion Forums
Discussions		 Project Information
Project Info
Filesystems/Discussion/General/QNX 6.4 - QNX 6.3 FS Comparison/List of Posts
Forum Topic - QNX 6.4 - QNX 6.3 FS Comparison: (10 Items)
   
Update
Juan Manuel Placco
11/20/2008 9:37 AM
post16968
QNX 6.4 - QNX 6.3 FS Comparison   
Hi,

As Robert suggest I redirect my bulk of file-system comparison (fs-qnx4 vs fs-qnx6) questions to this forum. In this 
link you'll find a modest pps with my questions (Comparing QNX 6.4 vs QNX 6.3 Forum v1.pps )

http://community.qnx.com/sf/go/projects.networking/discussion.technology.topc5048 

I will be very thankful if they can help me with this subject!

Thanks in advance!

Regards,
Juan Manuel
Juan Manuel Placco
11/21/2008 2:40 PM
post17088
Re: QNX 6.4 - QNX 6.3 FS Comparison   
Any news?
Juan Manuel Placco
11/22/2008 1:27 PM
post17132
Re: QNX 6.4 - QNX 6.3 FS Comparison   
Here I post this little presentation with some filesystem comparison between qnx4-fs and qnx6-fs.

I would be very thankful if they could clarify me these doubts.

Regards,
Juan Manuel
Attachment: Powerpoint Comparing FS QNX 6.4 vs QNX 6.3 Forum v1.pps 686 KB
David Sarrazin
11/23/2008 12:52 AM
post17138
Re: QNX 6.4 - QNX 6.3 FS Comparison   
Juan,

You don't show the partition types for each machine (6.3 vs 6.4), but you do say "standard installation".  I assume then
 that the 6.4 machine is using partition type 179, and the 6.3 machine is using partition type 79.  The 6.4 release uses
 a new filesystem, called the QNX6 filesystem.  All releases of QNX Neutrino before 6.4 used the QNX4 filesystem.

The new QNX6 filesystem uses copy-on-write to ensure data integrity.  This new scheme is much safer, and allows for much
 faster startup times, especially if the power is suddenly lost during filesystem writes.  The trade-off is that the new
 QNX6 filesystem must do additional writing, to implement the copy-on-write.  This can result in slightly lower write 
throughput as compared to the QNX4 filesystem.

If you wish to compare 6.3 versus 6.4, using the same filesystem, the installer offers a choice when partitioning the 
HDD, if you choose the "power-safe fillesystem" that is the new QNX6 FS.  Choose partition type 79 to get the older  
QNX4 filesystem that's used in 6.3.2.


More detail on the new QNX6 filesystem can be found at this link.

http://community.qnx.com/sf/wiki/do/viewPage/projects.filesystems/wiki/QNX6CopyOnWrite
Juan Manuel Placco
11/23/2008 12:19 PM
post17142
Re: QNX 6.4 - QNX 6.3 FS Comparison   
David, thank you very much for the response! 

In a general way, I supposed that it was going to exist such performance difference on 79 vs 179 partition types because
 of the implementation on the copy-on-write as you mention, after reading the wiki introduction. Mostly in my 'byte per 
byte' throughput test. 
(>> WIKI: The first reason is that most people do not write a single byte at a time to a block device)

Other thing: you have mentioned:
>> The new QNX6 filesystem must do additional writing, to implement the copy-on-write This can result in slightly lower 
write throughput as compared to the QNX4 filesystem. 

According to my workbench's numbers in the 'file descriptor' method, the performance was decremented in aprox 73% in 
both cases (dev1 and dev2), which I found too much. 

It would be very interesting for me to know why this huge difference (in 'fd' method) in both OS. It's only about the FS
 implementation? Or the procnto is implied? Since I call 'open' once at the beggining, not in every write or read. I 
will be very thankful if you can help me to clarify these.

The differences in FILE* testing method were more reasonable (15% slower in DEV2), and even faster (1%) in DEV1 node (2 
CPU with 1 GB RAM)

But what worries a little to me is how it will be the performance difference working with a data base (MySQL for 
example) in hard, real time conditions and how can I deal with that (using the power-safe qnx6 filesystem, of course)

Thank you very much again!

Regards,
Juan Manuel
Mario Charest
11/24/2008 8:22 AM
post17157
RE: QNX 6.4 - QNX 6.3 FS Comparison   
How bit are the read/writes. My obversation is that over 4K fd operation becomes faster then FILE* operation. That is to
 be expected because FILE* is build on TOP of FD operation.  If the size of the operation is to small you aren`t just 
testing the filesystem but also message passing and context switch.

-----Original Message-----
From: Juan Manuel Placco [mailto:community-noreply@qnx.com] 
Sent: November-22-08 1:28 PM
To: general-filesystems
Subject: Re: QNX 6.4 - QNX 6.3 FS Comparison

Here I post this little presentation with some filesystem comparison between qnx4-fs and qnx6-fs.

I would be very thankful if they could clarify me these doubts.

Regards,
Juan Manuel


_______________________________________________
General
http://community.qnx.com/sf/go/post17132
Juan Manuel Placco
11/24/2008 8:54 AM
post17159
Re: RE: QNX 6.4 - QNX 6.3 FS Comparison   
> How bit are the read/writes. My obversation is that over 4K fd operation 
> becomes faster then FILE* operation. That is to be expected because FILE* is 
> build on TOP of FD operation.  If the size of the operation is to small you 
> aren`t just testing the filesystem but also message passing and context switch
> .
> 
> -----Original Message-----
> From: Juan Manuel Placco [mailto:community-noreply@qnx.com] 
> Sent: November-22-08 1:28 PM
> To: general-filesystems
> Subject: Re: QNX 6.4 - QNX 6.3 FS Comparison
> 
> Here I post this little presentation with some filesystem comparison between 
> qnx4-fs and qnx6-fs.
> 
> I would be very thankful if they could clarify me these doubts.
> 
> Regards,
> Juan Manuel
> 
> 
> _______________________________________________
> General
> http://community.qnx.com/sf/go/post17132


Thanks Mario, I supposed that. That why I was aswering if procnto was implied too in the operation.

But still continuous with the doubt. Why 6.4 is much slower, now we could say, in message passing and context switch?

Thank you very much again!

Regards,
Juan Manuel
Mario Charest
11/24/2008 8:58 AM
post17163
RE: RE: QNX 6.4 - QNX 6.3 FS Comparison   

-----Original Message-----
From: Juan Manuel Placco [mailto:community-noreply@qnx.com] 
Sent: November-24-08 8:54 AM
To: general-filesystems
Subject: Re: RE: QNX 6.4 - QNX 6.3 FS Comparison

> How bit are the read/writes. My obversation is that over 4K fd operation 
> becomes faster then FILE* operation. That is to be expected because FILE* is 
> build on TOP of FD operation.  If the size of the operation is to small you 
> aren`t just testing the filesystem but also message passing and context switch
> .
> 
> -----Original Message-----
> From: Juan Manuel Placco [mailto:community-noreply@qnx.com] 
> Sent: November-22-08 1:28 PM
> To: general-filesystems
> Subject: Re: QNX 6.4 - QNX 6.3 FS Comparison
> 
> Here I post this little presentation with some filesystem comparison between 
> qnx4-fs and qnx6-fs.
> 
> I would be very thankful if they could clarify me these doubts.
> 
> Regards,
> Juan Manuel
> 
> 
> _______________________________________________
> General
> http://community.qnx.com/sf/go/post17132


Thanks Mario, I supposed that. That why I was aswering if procnto was implied too in the operation.

But still continuous with the doubt. Why 6.4 is much slower, now we could say, in message passing and context switch?

An educated guess is that the 6.4 kernel has more feature to support then 4.xx hence it might be slower. One thing that 
you haven't check though in your benchmark is CPU usage.  That's an important factor to consider.

If only QSS would release the beta of Deja-View for QNX4 ( it's a system profiler ) you could make are very deep 
analysis of what component is faster/slower between the two OSes.

Thank you very much again!

Regards,
Juan Manuel


_______________________________________________
General
http://community.qnx.com/sf/go/post17159
Juan Manuel Placco
11/26/2008 10:51 AM
post17383
Re: RE: RE: QNX 6.4 - QNX 6.3 FS Comparison   
Mario, thanks for the reply.

>> One thing that you haven't check though in your benchmark 
>> is CPU usage.  That's an important factor to consider.

Here it is. While writing and reading a 2GB file byte per byte (files1)

hogs -n -s1 -%2
1                      896   43%    89%
8200       devb-eide   201    9%    20%
1191974       files1   799   39%    79%

1                     1061   51%   106%
8200       devb-eide   142    6%    14%
1191974       files1   681   33%    68%

1                     1017   49%   101%
8200       devb-eide   150    7%    15%
1191974       files1   823   40%    82%

1                      992   48%    99%
8200       devb-eide   178    8%    17%
1191974       files1   831   40%    83%

1                      984   48%    98%
8200       devb-eide   171    8%    17%
1191974       files1   842   41%    84%

# pidin -d10 -l sched | grep file
 1191974   1 ./files1            10r   1                      REPLY
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING

It seems the average CPU usage is about 35-40% and the 2 CPUs are working (there are no other hogging processes running)
. What does this numbers means to my per-byte W/R test?

>> If only QSS would release the beta of Deja-View for QNX4 
>>( it's a system profiler ) you could make are very deep analysis of 
>> what component is faster/ slower between the two OSes.

>> ... the 6.4 kernel has more feature to support then 4.xx 
>> hence it might be slower

Sorry Mario, we're talking about 6.3 & 6.4, aren't we?

Thank you very much!

Regards,
Juan Manuel
Mario Charest
11/26/2008 10:58 AM
post17386
RE: RE: RE: QNX 6.4 - QNX 6.3 FS Comparison   
I haven't look at your code ( your posted it ? ) but I find it strange the files1 is using all that CPU.  If the program
 is only reading/writing it should use very little CPU time.  CPU is not 35%-40% but more like 50% which means one core 
is use 100% of the time. That's not typical for a filesystem benchmark.

Yes I'm talking about 6.X kernel versus 4.2X

-----Original Message-----
From: Juan Manuel Placco [mailto:community-noreply@qnx.com] 
Sent: November-26-08 10:51 AM
To: general-filesystems
Subject: Re: RE: RE: QNX 6.4 - QNX 6.3 FS Comparison

Mario, thanks for the reply.

>> One thing that you haven't check though in your benchmark 
>> is CPU usage.  That's an important factor to consider.

Here it is. While writing and reading a 2GB file byte per byte (files1)

hogs -n -s1 -%2
1                      896   43%    89%
8200       devb-eide   201    9%    20%
1191974       files1   799   39%    79%

1                     1061   51%   106%
8200       devb-eide   142    6%    14%
1191974       files1   681   33%    68%

1                     1017   49%   101%
8200       devb-eide   150    7%    15%
1191974       files1   823   40%    82%

1                      992   48%    99%
8200       devb-eide   178    8%    17%
1191974       files1   831   40%    83%

1                      984   48%    98%
8200       devb-eide   171    8%    17%
1191974       files1   842   41%    84%

# pidin -d10 -l sched | grep file
 1191974   1 ./files1            10r   1                      REPLY
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING
 1191974   1 ./files1            10r   0                      RUNNING
 1191974   1 ./files1            10r   1                      RUNNING

It seems the average CPU usage is about 35-40% and the 2 CPUs are working (there are no other hogging processes running)
. What does this numbers means to my per-byte W/R test?

>> If only QSS would release the beta of Deja-View for QNX4 
>>( it's a system profiler ) you could make are very deep analysis of 
>> what component is faster/ slower between the two OSes.

>> ... the 6.4 kernel has more feature to support then 4.xx 
>> hence it might be slower

Sorry Mario, we're talking about 6.3 & 6.4, aren't we?

Thank you very much!

Regards,
Juan Manuel


_______________________________________________
General
http://community.qnx.com/sf/go/post17383

Return


  A subsidiary of BlackBerry             All content ©2004-2013, QNX Software Systems