Memory allocation speed is slow by memory fragmentation on long-run test.

JunBeom Kim jbkim at e-coretech.kr
Fri Mar 29 07:47:43 UTC 2019


Dear Sebastian,

There is not any problem using my patch code in my customer side until now.
I will consider TLSF.

Thank you very much for TLSF information.

On referencing, I will investigate libbsd's SD-Card speed improvement for
IMX target(your side: i.MX7D, my side: i.MX6Q) soon.

If there is internal progress, I will share patch code for you.

Best Regards,
JunBeom

-----Original Message-----
From: Sebastian Huber <sebastian.huber at embedded-brains.de> 
Sent: Friday, March 29, 2019 4:39 PM
To: JunBeom Kim <jbkim at e-coretech.kr>; users at rtems.org
Subject: Re: Memory allocation speed is slow by memory fragmentation on
long-run test.



On 29/03/2019 08:19, JunBeom Kim wrote:
>
> Dear Sir,
>
> I am still working with my customer using RTEMS based Qt Framework in 
> long time.
>
> My customer are using memory allocation(malloc, realloc, free) with 
> frequency about small size(64, 128, 256, 512, 1024 byte, etc).
>
> When RTEMS application is run in long time(for example: 6 hours),
> malloc() or realloc() speed is slower due to memory fragmentation.
>
> When I used other RTOS(eg: Nucleus, uC/OS, iTRON, etc) before, there 
> was almost same issue.
>

A first-fit allocator (which is used by RTEMS) is completely useless for
dynamic memory workloads. It is all right if you only allocate and never
free.

> I used partition memory management for resolving this.
>
> Is it correct method ?
>

We used TLSF in such situations:

https://github.com/mattconte/tlsf

--
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone   : +49 89 189 47 41-16
Fax     : +49 89 189 47 41-09
E-Mail  : sebastian.huber at embedded-brains.de
PGP     : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.



More information about the users mailing list