ppc405ex + gigabit ethernet

ppc405ex + gigabit ethernet

[Lada Podivin]

[-- Attachment #1: Type: text/plain, Size: 635 bytes --]

Hi,

I benchmarked performance of my network, which contains ppc405EX (Kilauea
board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit
ethernet. I used the netperf tool:

netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768

So I was sending UDP packets to the desktop. The resulting speed was about
370 Kb/s. I tried to send the packets to several different computers - with
the same result. So the ppc board is the bottleneck in this case.

Is there any possibility to improve the gigabit capabilities of the
ppc405EX? Is there anyone who achieved a better performance with ppc4xx
boards?

Thanks!

Best,
Lada Podivin

[-- Attachment #2: Type: text/html, Size: 685 bytes --]

RE: ppc405ex + gigabit ethernet

[LiuMing]

[-- Attachment #1: Type: text/plain, Size: 1290 bytes --]


Hi,

According to our experience on PPC405 + Gigabit Enet, your result is already very reasonable. For UDP transmission, it can be around 400 Kb/s because of the CPU bottleneck (in our case 300 MHz ppc405). If you want to further improve it, a faster CPU is needed to process the TCP/IP stack, or you may buy commercial TCP/IP stack rather than using the free Linux one. 

 

BR

Ming
 


Date: Tue, 30 Jun 2009 15:58:53 +0200
Subject: ppc405ex + gigabit ethernet
From: lada.podivin@gmail.com
To: linuxppc-dev@ozlabs.org

Hi,

I benchmarked performance of my network, which contains ppc405EX (Kilauea board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit ethernet. I used the netperf tool: 

netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768

So I was sending UDP packets to the desktop. The resulting speed was about 370 Kb/s. I tried to send the packets to several different computers - with the same result. So the ppc board is the bottleneck in this case.

Is there any possibility to improve the gigabit capabilities of the ppc405EX? Is there anyone who achieved a better performance with ppc4xx boards?

Thanks!

Best,
Lada Podivin


_________________________________________________________________
MSN 表情魔法书，改变你的对话时代！
http://im.live.cn/emoticons/

[-- Attachment #2: Type: text/html, Size: 1580 bytes --]

Re: ppc405ex + gigabit ethernet

[Lada Podivin]

[-- Attachment #1: Type: text/plain, Size: 2167 bytes --]

Thank you for your reply!
Yes, I agree the CPU is the bottleneck. But I have performed more tests with
the netperf tool and results seem strange to me. If I send 800 B of data,
the resulting speed is aproximately 100 Kb/s. On the other hand, If I try to
send 32KB chunk of data with jumbo frames turned on, resulting speed is
about 600 Kb/s (300-400 without jumbo frames). My ppc405ex runs at 600 MHz.

So it seems that it is definetely better idea to call sendto() less
frequently with bigger amounts of data, than to call it more often with
smaler amounts. Is it because smaller data means more frequent user
space/kernel space switching? Or is it network related issue?

Thanks!

Best,
Lada

2009/7/1 LiuMing <eemingliu@hotmail.com>

>  Hi,
> According to our experience on PPC405 + Gigabit Enet, your result is
> already very reasonable. For UDP transmission, it can be around 400 Kb/s
> because of the CPU bottleneck (in our case 300 MHz ppc405). If you want
> to further improve it, a faster CPU is needed to process the TCP/IP stack,
> or you may buy commercial TCP/IP stack rather than using the free Linux one.
>
>
> BR
> Ming
>
> ------------------------------
> Date: Tue, 30 Jun 2009 15:58:53 +0200
> Subject: ppc405ex + gigabit ethernet
> From: lada.podivin@gmail.com
> To: linuxppc-dev@ozlabs.org
>
>
> Hi,
>
> I benchmarked performance of my network, which contains ppc405EX (Kilauea
> board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit
> ethernet. I used the netperf tool:
>
> netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768
>
> So I was sending UDP packets to the desktop. The resulting speed was about
> 370 Kb/s. I tried to send the packets to several different computers - with
> the same result. So the ppc board is the bottleneck in this case.
>
> Is there any possibility to improve the gigabit capabilities of the
> ppc405EX? Is there anyone who achieved a better performance with ppc4xx
> boards?
>
> Thanks!
>
> Best,
> Lada Podivin
>
>
> ------------------------------
> 立刻下载 MSN 保护盾，保障 MSN 安全稳定！ 现在就下载！ <http://im.live.cn/safe/>
>

[-- Attachment #2: Type: text/html, Size: 2645 bytes --]

Re: ppc405ex + gigabit ethernet

[Detlev Zundel]

Hi Lada,

> Hi,
>
> I benchmarked performance of my network, which contains ppc405EX (Kilauea
> board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit
> ethernet. I used the netperf tool:
>
> netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768
>
> So I was sending UDP packets to the desktop. The resulting speed was about 370
> Kb/s. I tried to send the packets to several different computers - with the
> same result. So the ppc board is the bottleneck in this case.
>
> Is there any possibility to improve the gigabit capabilities of the ppc405EX?
> Is there anyone who achieved a better performance with ppc4xx boards?

On our kilauea in the lab:

-bash-3.2# src/netperf -t UDP_STREAM -p 7776 -H 192.168.1.1 -fK -- -m 32768
UDP UNIDIRECTIONAL SEND TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.1.1 (192.168.1.1) port 0 AF_INET
Socket  Message  Elapsed      Messages                
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   KBytes/sec

106496   32768   10.00        3601      0    11519.64
124928           10.00        3601           11519.64

-bash-3.2# grep cpu /proc/cpuinfo 
cpu             : 405EX
-bash-3.2# cat /proc/version
Linux version 2.6.29.4 (dzu@pollux.denx.de) (gcc version 4.2.2) #9 Wed Jun 17 11:18:46 CEST 2009
-bash-3.2# 

I can send you the kernel+dtb that I used for testing offlist.

Cheers
  Detlev

--
DENX Software Engineering GmbH,      MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich,  Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-40 Fax: (+49)-8142-66989-80 Email: dzu@denx.de

Re: ppc405ex + gigabit ethernet

[Lada Podivin]

[-- Attachment #1: Type: text/plain, Size: 2337 bytes --]

Eh! I mean Mb/s NOT Kb/s! Sorry!

2009/7/1 Lada Podivin <lada.podivin@gmail.com>

> Thank you for your reply!
> Yes, I agree the CPU is the bottleneck. But I have performed more tests
> with the netperf tool and results seem strange to me. If I send 800 B of
> data, the resulting speed is aproximately 100 Kb/s. On the other hand, If I
> try to send 32KB chunk of data with jumbo frames turned on, resulting speed
> is about 600 Kb/s (300-400 without jumbo frames). My ppc405ex runs at 600
> MHz.
>
> So it seems that it is definetely better idea to call sendto() less
> frequently with bigger amounts of data, than to call it more often with
> smaler amounts. Is it because smaller data means more frequent user
> space/kernel space switching? Or is it network related issue?
>
> Thanks!
>
> Best,
> Lada
>
> 2009/7/1 LiuMing <eemingliu@hotmail.com>
>
>  Hi,
>> According to our experience on PPC405 + Gigabit Enet, your result is
>> already very reasonable. For UDP transmission, it can be around 400 Kb/s
>> because of the CPU bottleneck (in our case 300 MHz ppc405). If you want
>> to further improve it, a faster CPU is needed to process the TCP/IP stack,
>> or you may buy commercial TCP/IP stack rather than using the free Linux one.
>>
>>
>> BR
>> Ming
>>
>> ------------------------------
>> Date: Tue, 30 Jun 2009 15:58:53 +0200
>> Subject: ppc405ex + gigabit ethernet
>> From: lada.podivin@gmail.com
>> To: linuxppc-dev@ozlabs.org
>>
>>
>> Hi,
>>
>> I benchmarked performance of my network, which contains ppc405EX (Kilauea
>> board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit
>> ethernet. I used the netperf tool:
>>
>> netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768
>>
>> So I was sending UDP packets to the desktop. The resulting speed was about
>> 370 Kb/s. I tried to send the packets to several different computers - with
>> the same result. So the ppc board is the bottleneck in this case.
>>
>> Is there any possibility to improve the gigabit capabilities of the
>> ppc405EX? Is there anyone who achieved a better performance with ppc4xx
>> boards?
>>
>> Thanks!
>>
>> Best,
>> Lada Podivin
>>
>>
>> ------------------------------
>> 立刻下载 MSN 保护盾，保障 MSN 安全稳定！ 现在就下载！ <http://im.live.cn/safe/>
>>
>
>

[-- Attachment #2: Type: text/html, Size: 3043 bytes --]

RE: ppc405ex + gigabit ethernet

[Cote, Sylvain]

[-- Attachment #1: Type: text/plain, Size: 2017 bytes --]




> Hi,

> I benchmarked performance of my network, which contains ppc405EX (Kilauea board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit ethernet. I used the > netperf tool:

> netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768

> So I was sending UDP packets to the desktop. The resulting speed was about 370 Kb/s. I tried to send the packets to several different computers - with the same result. So the >ppc board is the bottleneck in this case.

> Is there any possibility to improve the gigabit capabilities of the ppc405EX? Is there anyone who achieved a better performance with ppc4xx boards?
We have seen also bad performances with Gig Ethernet with ppc405ex.  One optimisation we have done is interrupt coalescing in ibm_new_mac.  This gives near 25% of txput gain.  Unfortunately, this optimisation is not yet on denx git.  We got it from AMCC directly.  Also, it looks like that 2.6 kernel is much slower then 2.4 on ppc405ex.  This is probably due to IPC and/or context switching.  We have not yet found a solution to this.  However, 370Kb/s (3Mbits/s) look like very poor.  In the worst case with 100% CPU used to send UDP data on network with the ppc405ex 600MHz, we got 190Mbits/s on Gig eth.  It looks like netperf does not stress the ppc405ex ant it network at is maximum.

Sylvain



________________________________
立刻下载 MSN 保护盾，保障 MSN 安全稳定！ 现在就下载！<http://im.live.cn/safe/>

This electronic message may contain proprietary and confidential information of Verint Systems Inc., its affiliates and/or subsidiaries.
The information is intended to be for the use of the individual(s) or
entity(ies) named above.  If you are not the intended recipient (or authorized to receive this e-mail for the intended recipient), you may not use, copy, disclose or distribute to anyone this message or any information contained in this message.  If you have received this electronic message in error, please notify us by replying to this e-mail.


[-- Attachment #2: Type: text/html, Size: 6498 bytes --]

RE: ppc405ex + gigabit ethernet

[john.p.price]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="GB2312", Size: 2649 bytes --]

I have developed a board with the AMCC PPC405eX running at 600MHz and experimented with two transmission methods;

 

1.    Raw socket w/jumbo frames ¨C approx. 600Mbits/sec

2.    TCP using writev w/jumbo frames ¨C approx. 525Mbits/sec

 

From: linuxppc-dev-bounces+john.p.price=l-3com.com@lists.ozlabs.org [mailto:linuxppc-dev-bounces+john.p.price=l-3com.com@lists.ozlabs.org] On Behalf Of Lada Podivin
Sent: Wednesday, July 01, 2009 7:25 AM
To: LiuMing
Cc: linuxppc-dev@ozlabs.org
Subject: Re: ppc405ex + gigabit ethernet

 

Thank you for your reply!
Yes, I agree the CPU is the bottleneck. But I have performed more tests with the netperf tool and results seem strange to me. If I send 800 B of data, the resulting speed is aproximately 100 Kb/s. On the other hand, If I try to send 32KB chunk of data with jumbo frames turned on, resulting speed is about 600 Kb/s (300-400 without jumbo frames). My ppc405ex runs at 600 MHz.

So it seems that it is definetely better idea to call sendto() less frequently with bigger amounts of data, than to call it more often with smaler amounts. Is it because smaller data means more frequent user space/kernel space switching? Or is it network related issue?  

Thanks!

Best,
Lada

2009/7/1 LiuMing <eemingliu@hotmail.com>

Hi,
According to our experience on PPC405 + Gigabit Enet, your result is already very reasonable. For UDP transmission, it can be around 400 Kb/s because of the CPU bottleneck (in our case 300 MHz ppc405). If you want to further improve it, a faster CPU is needed to process the TCP/IP stack, or you may buy commercial TCP/IP stack rather than using the free Linux one. 
 
BR
Ming
 

________________________________

Date: Tue, 30 Jun 2009 15:58:53 +0200
Subject: ppc405ex + gigabit ethernet
From: lada.podivin@gmail.com
To: linuxppc-dev@ozlabs.org



Hi,

I benchmarked performance of my network, which contains ppc405EX (Kilauea board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit ethernet. I used the netperf tool: 

netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768

So I was sending UDP packets to the desktop. The resulting speed was about 370 Kb/s. I tried to send the packets to several different computers - with the same result. So the ppc board is the bottleneck in this case.

Is there any possibility to improve the gigabit capabilities of the ppc405EX? Is there anyone who achieved a better performance with ppc4xx boards?

Thanks!

Best,
Lada Podivin



________________________________

Á¢¿ÌÏÂÔØ MSN ±£»¤¶Ü£¬±£ÕÏ MSN °²È«Îȶ¨£¡ ÏÖÔÚ¾ÍÏÂÔØ£¡ <http://im.live.cn/safe/> 

 


[-- Attachment #2: Type: text/html, Size: 10727 bytes --]

RE: ppc405ex + gigabit ethernet

[LiuMing]

[-- Attachment #1: Type: text/plain, Size: 2915 bytes --]


Yes. If you enable jumbo frames, the results will be definitely better. My result is with jumbo frame of 8982 enabled. 

 

If you say your CPU is 600 MHz, I guess there is still potential for you to improve the speed. In our case we had only 300 MHz and got a similar value of yours. Of course, there are many other issues which can be enabled to improve the performance, such as interrupt coaleasing, hardware checksum processing, etc.. Since we used Xilinx FPGA with embedded PowerPC and HW TEMAC, some work can be done with FPGA hardware resources. I don't know your case and it's hard to say how can you further improve. 

 

BR

Ming
 


Date: Wed, 1 Jul 2009 13:24:41 +0200
Subject: Re: ppc405ex + gigabit ethernet
From: lada.podivin@gmail.com
To: eemingliu@hotmail.com
CC: linuxppc-dev@ozlabs.org

Thank you for your reply!
Yes, I agree the CPU is the bottleneck. But I have performed more tests with the netperf tool and results seem strange to me. If I send 800 B of data, the resulting speed is aproximately 100 Kb/s. On the other hand, If I try to send 32KB chunk of data with jumbo frames turned on, resulting speed is about 600 Kb/s (300-400 without jumbo frames). My ppc405ex runs at 600 MHz.

So it seems that it is definetely better idea to call sendto() less frequently with bigger amounts of data, than to call it more often with smaler amounts. Is it because smaller data means more frequent user space/kernel space switching? Or is it network related issue?  

Thanks!

Best,
Lada


2009/7/1 LiuMing <eemingliu@hotmail.com>


Hi,
According to our experience on PPC405 + Gigabit Enet, your result is already very reasonable. For UDP transmission, it can be around 400 Kb/s because of the CPU bottleneck (in our case 300 MHz ppc405). If you want to further improve it, a faster CPU is needed to process the TCP/IP stack, or you may buy commercial TCP/IP stack rather than using the free Linux one. 
 
BR
Ming
 


Date: Tue, 30 Jun 2009 15:58:53 +0200
Subject: ppc405ex + gigabit ethernet
From: lada.podivin@gmail.com
To: linuxppc-dev@ozlabs.org




Hi,

I benchmarked performance of my network, which contains ppc405EX (Kilauea board, kernel 2.6.30 from Denx) connected with a linux desktop via gigabit ethernet. I used the netperf tool: 

netperf -t UDP_STREAM -H 192.168.1.1 -- -m 32768

So I was sending UDP packets to the desktop. The resulting speed was about 370 Kb/s. I tried to send the packets to several different computers - with the same result. So the ppc board is the bottleneck in this case.

Is there any possibility to improve the gigabit capabilities of the ppc405EX? Is there anyone who achieved a better performance with ppc4xx boards?

Thanks!

Best,
Lada Podivin





立刻下载 MSN 保护盾，保障 MSN 安全稳定！ 现在就下载！

_________________________________________________________________
Messenger安全保护中心，免费修复系统漏洞，保护Messenger安全！
http://im.live.cn/safe/

[-- Attachment #2: Type: text/html, Size: 3699 bytes --]

Re: ppc405ex + gigabit ethernet

[Lada Podivin]

[-- Attachment #1: Type: text/plain, Size: 64 bytes --]

Many thanks for all responses! Now I know all I need to know :)

[-- Attachment #2: Type: text/html, Size: 71 bytes --]

Re: ppc405ex + gigabit ethernet

[Lada Podivin]

Hi Sylvain,

the interrupt coalescing sounds like good idea - I'm surprised this
feature is missing in the original ibm_newemac driver. You wrote you
had got this optimisation directly from AMCC. Is it part of any
framework? I'm just wondering how one can obtain it. I tried to find
any suitable patch but with no success - the old friend Google didn't
help this time :)

Thank you very much!
Lada

RE: ppc405ex + gigabit ethernet

[Feng Kan]

[-- Attachment #1: Type: text/plain, Size: 922 bytes --]

Hi Lada:

Please contact support@amcc.com for additional help for the coalescing patch.

Feng Kan
AMCC Software


-----Original Message-----
From: linuxppc-dev-bounces+fkan=amcc.com@lists.ozlabs.org on behalf of Lada Podivin
Sent: Fri 7/3/2009 2:09 AM
To: Cote, Sylvain
Cc: linuxppc-dev@ozlabs.org
Subject: Re: ppc405ex + gigabit ethernet
 
Hi Sylvain,

the interrupt coalescing sounds like good idea - I'm surprised this
feature is missing in the original ibm_newemac driver. You wrote you
had got this optimisation directly from AMCC. Is it part of any
framework? I'm just wondering how one can obtain it. I tried to find
any suitable patch but with no success - the old friend Google didn't
help this time :)

Thank you very much!
Lada
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev


[-- Attachment #2: Type: text/html, Size: 1480 bytes --]