udt-general — UDT General Discussion

Re: [UDT] UDT with congestion control

[Harald s. <sc...@rv...>]

Hello Yunhong,

Thank you for your answer.
I have also tried the scalable, hstcp and BIC congestion contol
algorithms as they are defined in cc.h.
Every time i get a performance of 0.2 MB/s.

For example, i activated scalable with:
UDT::setsockopt(client,0,UDT_CC,newCCCFactory<CScalableTCP>,sizeof(CCCFac=
tory<CScalableTCP>));
Is there anything more to do? What may be the reason for the poor
Performance?

Thanks in advance,
Harald


> Hi, Harald,
>=20
> Once you uncomment the -DCUSTOM_CC option, the default UDT congestion=20
> control algorithm will not be used any more.
>=20
> Default window size in UDT is 16 packets and it will not change unless =
you=20
> assign a new congestion control algorithm.
>=20
> Once you specify CTCP, you are using TCP's AIMD algorithm, which usuall=
y=20
> does not work well for long distance links (you mentioned that RTT is=20
> 100ms).
>=20
> Thanks
> Yunhong
>=20
>=20
> On Thu, 2 Mar 2006, Harald schwier wrote:
>=20
> > Hello,
> > i'm doing some performance tests with UDT. Without changing
> > any options, appclient shows 950 MB/s (1GB Ethernet, RTT about 100ms).
> >
> > After i uncomment the -DCUSTOM_CC compile option in ./src/Makefile
> > and recompile the library the Performance goes down to 1.7 MB/s witho=
ut
> > changing any Options in appclient.
> >
> > When i change the socket options with:
> > UDT::setsockopt(client,0,UDT_CC,newCCCFactory<CTCP>,sizeof(CCCFactory=
<CTCP>));
> > the performance goes down to 0.2 MB/s.
> >
> > UDT 2.3
> > Compiled with: make -e os=3DLINUX arch=3DAMD64
> >
> > What may be the reason for the poor performance???
> >
> > Thanks in advance,
> > Harald
> >
> >
> >
> >
> > -------------------------------------------------------
> > This SF.Net email is sponsored by xPML, a groundbreaking scripting la=
nguage
> > that extends applications into web and mobile media. Attend the live =
webcast
> > and join the prime developer group breaking into this new coding terr=
itory!
> > http://sel.as-us.falkag.net/sel?cmd=3Dlnk&kid=3D110944&bid=3D241720&d=
at=3D121642
> > _______________________________________________
> > UDT-general mailing list
> > UDT...@li...
> > https://lists.sourceforge.net/lists/listinfo/udt-general
> >
Mit freundlichen Gr=C3=BC=C3=9Fen,
Harald Schwier


--=20
Dipl.-Ing. Harald Schwier          D-30159 Hannover
Lehrgebiet Rechnernetze            Schlosswender Str. 5
RRZN, Universit=C3=A4t Hannover         Tel: +49 (511) 762 7919081
Fachbereich Informatik

Re: [UDT] UDT with congestion control

[Yunhong Gu <yu...@la...>]

Hi, Harald,

Once you uncomment the -DCUSTOM_CC option, the default UDT congestion 
control algorithm will not be used any more.

Default window size in UDT is 16 packets and it will not change unless you 
assign a new congestion control algorithm.

Once you specify CTCP, you are using TCP's AIMD algorithm, which usually 
does not work well for long distance links (you mentioned that RTT is 
100ms).

Thanks
Yunhong


On Thu, 2 Mar 2006, Harald schwier wrote:

> Hello,
> i'm doing some performance tests with UDT. Without changing
> any options, appclient shows 950 MB/s (1GB Ethernet, RTT about 100ms).
>
> After i uncomment the -DCUSTOM_CC compile option in ./src/Makefile
> and recompile the library the Performance goes down to 1.7 MB/s without
> changing any Options in appclient.
>
> When i change the socket options with:
> UDT::setsockopt(client,0,UDT_CC,newCCCFactory<CTCP>,sizeof(CCCFactory<CTCP>));
> the performance goes down to 0.2 MB/s.
>
> UDT 2.3
> Compiled with: make -e os=LINUX arch=AMD64
>
> What may be the reason for the poor performance???
>
> Thanks in advance,
> Harald
>
>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by xPML, a groundbreaking scripting language
> that extends applications into web and mobile media. Attend the live webcast
> and join the prime developer group breaking into this new coding territory!
> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642
> _______________________________________________
> UDT-general mailing list
> UDT...@li...
> https://lists.sourceforge.net/lists/listinfo/udt-general
>

[UDT] UDT with congestion control

[Harald s. <sc...@rv...>]

Hello,
i'm doing some performance tests with UDT. Without changing
any options, appclient shows 950 MB/s (1GB Ethernet, RTT about 100ms).

After i uncomment the -DCUSTOM_CC compile option in ./src/Makefile
and recompile the library the Performance goes down to 1.7 MB/s without
changing any Options in appclient.

When i change the socket options with:
UDT::setsockopt(client,0,UDT_CC,newCCCFactory<CTCP>,sizeof(CCCFactory<CTCP>));
the performance goes down to 0.2 MB/s.

UDT 2.3
Compiled with: make -e os=LINUX arch=AMD64

What may be the reason for the poor performance???

Thanks in advance,
Harald

Re: [UDT] UDT Resource Usage

[Ben P. <ben...@ra...>]

Hi,

Thanks for getting back to me. The increased buffer size certainly 
explains the differences in memory usage.

Unfortunately I haven't had time to further investigate the CPU usage 
yet although I can confirm that I was using 2.2.

Thanks a lot for your help,

Ben Page


Yunhong Gu wrote:
> 
> 
> Hi, Ben,
> 
> As you know that UDT was designed for high speed networks, the default 
> protocol buffer size (40MB) is much larger than the common system socket 
> settings. You can use "UDT::setsockopt" to modify the protocol buffer size.
> 
> E.g.,
> UDT::setsockopt(client, 0, UDP_RCVBUF, new int(1000000), sizeof(int));
> 
> When NO_BUSY_WAITING is enabled, UDT should not use much CPU time, so 
> that was strange. Are you using version 2.2 or 2.1? If you are using 
> 2.1, can you give a try of 2.2? I just tested it on my Windows XP 
> machine, and it costed about 2% CPU when sending data at 100Mb/s.
> 
> Thanks,
> Gu
> 
> Yunhong Gu, Research Scientist
> 
> National Center for Data Mining
> University of Illinois at Chicago
> SEO 700, M/C 249, 851 S Morgan St
> Chicago, IL 60607-7045
> T (312) 413-9576
> F (312) 355-0373
> 
> On Tue, 17 Jan 2006, Ben Page wrote:
> 
>> Hello,
>>
>> I am currently evaluating the use of UDT for a new internet-based 
>> application I am working on. I have had good success integrating your 
>> library with my code but have noticed that when running the new 
>> version of my application with UDT the CPU usage jumps to near 100% 
>> and the memory usage is also greatly increased.
>>
>> While I understand that UDT is designed for high-bandwidth wide-area
>> networks and mine is an Internet-based application I would be interested
>> in any insight you might have into the highly resource-intensive
>> performance I have noticed.
>>
>> I am currently building UDT using the Visual Studio solution file
>> available with the latest release ( in which the NO_BUSY_WAIT flag is
>> set). I am using a statically-linked version of the library but apart
>> from that have made no changes to the build settings.
>>
>> The problem is more obvious when sending data than receiving. I'd be
>> really grateful if you could provide any insight into what I could be
>> doing wrong or where I should look while trying to fix the problem.
>>
>> Very Best Regards,
>>
>> Ben Page
>>
>>
>> -------------------------------------------------------
>> This SF.net email is sponsored by: Splunk Inc. Do you grep through log 
>> files
>> for problems?  Stop!  Download the new AJAX search engine that makes
>> searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
>> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
>> _______________________________________________
>> UDT-general mailing list
>> UDT...@li...
>> https://lists.sourceforge.net/lists/listinfo/udt-general
>>


-- 
Ben Page
Software Engineer
Rawflow Inc | 2nd Floor | The Old Pump House | 19 Hooper Street | London 
E1 8BU | United Kingdom

Tel:		+44 (0)20 7183 2222
Mobile:		+44 (0)7985 938 682
Fax:		+44 (0)20 7481 4343
URL:		www.rawflow.com

Re: [UDT] UDT Resource Usage

[Yunhong Gu <yu...@la...>]

Hi, Ben,

As you know that UDT was designed for high speed networks, the default 
protocol buffer size (40MB) is much larger than the common system socket 
settings. You can use "UDT::setsockopt" to modify the protocol buffer 
size.

E.g.,
UDT::setsockopt(client, 0, UDP_RCVBUF, new int(1000000), sizeof(int));

When NO_BUSY_WAITING is enabled, UDT should not use much CPU time, so that 
was strange. Are you using version 2.2 or 2.1? If you are using 2.1, can 
you give a try of 2.2? I just tested it on my Windows XP machine, and it 
costed about 2% CPU when sending data at 100Mb/s.

Thanks,
Gu

Yunhong Gu, Research Scientist

National Center for Data Mining
University of Illinois at Chicago
SEO 700, M/C 249, 851 S Morgan St
Chicago, IL 60607-7045
T (312) 413-9576
F (312) 355-0373

On Tue, 17 Jan 2006, Ben Page wrote:

> Hello,
>
> I am currently evaluating the use of UDT for a new internet-based application 
> I am working on. I have had good success integrating your library with my 
> code but have noticed that when running the new version of my application 
> with UDT the CPU usage jumps to near 100% and the memory usage is also 
> greatly increased.
>
> While I understand that UDT is designed for high-bandwidth wide-area
> networks and mine is an Internet-based application I would be interested
> in any insight you might have into the highly resource-intensive
> performance I have noticed.
>
> I am currently building UDT using the Visual Studio solution file
> available with the latest release ( in which the NO_BUSY_WAIT flag is
> set). I am using a statically-linked version of the library but apart
> from that have made no changes to the build settings.
>
> The problem is more obvious when sending data than receiving. I'd be
> really grateful if you could provide any insight into what I could be
> doing wrong or where I should look while trying to fix the problem.
>
> Very Best Regards,
>
> Ben Page
>
>
> -------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
> for problems?  Stop!  Download the new AJAX search engine that makes
> searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
> _______________________________________________
> UDT-general mailing list
> UDT...@li...
> https://lists.sourceforge.net/lists/listinfo/udt-general
>

[UDT] UDT Resource Usage

[Ben P. <ben...@ra...>]

Hello,

I am currently evaluating the use of UDT for a new internet-based 
application I am working on. I have had good success integrating your 
library with my code but have noticed that when running the new version 
of my application with UDT the CPU usage jumps to near 100% and the 
memory usage is also greatly increased.

While I understand that UDT is designed for high-bandwidth wide-area
networks and mine is an Internet-based application I would be interested
in any insight you might have into the highly resource-intensive
performance I have noticed.

I am currently building UDT using the Visual Studio solution file
available with the latest release ( in which the NO_BUSY_WAIT flag is
set). I am using a statically-linked version of the library but apart
from that have made no changes to the build settings.

The problem is more obvious when sending data than receiving. I'd be
really grateful if you could provide any insight into what I could be
doing wrong or where I should look while trying to fix the problem.

Very Best Regards,

Ben Page

[UDT] Re: UDT-general -- confirmation of subscription -- request 201343

[Luca De C. <lde...@gm...>]

On Tue, 2005-03-29 at 03:10 -0800,
udt...@li... wrote:
> UDT-general -- confirmation of subscription -- request 201343
> 
> We have received a request from 151.80.2.67 for subscription of your
> email address, <lde...@gm...>, to the
> udt...@li... mailing list.  To confirm the
> request, please send a message to
> udt...@li..., and either:
> 
> - maintain the subject line as is (the reply's additional "Re:" is
> ok),
> 
> - or include the following line - and only the following line - in the
> message body: 
> 
> confirm 201343
> 
> (Simply sending a 'reply' to this message should work from most email
> interfaces, since that usually leaves the subject line in the right
> form.)
> 
> If you do not wish to subscribe to this list, please simply disregard
> this message.  Send questions to
> udt...@li....