[NBLUG/talk] Samba running slow...

Coy Thorp C.Thorp at mdl.com
Tue Jul 15 13:17:00 PDT 2003


Ross,

 to answer your symmetry question.  Speed and duplex only affect the
connection from host to switch, not host to host (unless directly attached
by a cable).  If the speed setting is mismatched, then no connection,
bye-bye, thanks for playing (even though you may get a link light).  If
duplex is mismatched, then you will see transfers in one direction behave
normally, and transfers in another direction crawl.  A half duplex pipe pipe
does not expect to receive data during it's csma/cd time, so it becomes, how
you say, hosed. :)

-----Original Message-----
From: Todd Cary [mailto:todd at aristesoftware.com] 
Sent: Tuesday, July 15, 2003 12:27 PM
To: talk at nblug.org
Subject: Re: [NBLUG/talk] Samba running slow...


Ross -

To answer your question about how I measure the speed of the transfers, 
I use the Windows Explorer to move an 8 MB file.  It gives the rate and 
when it is running slow, the rate is ~500 Kbps (takes ~ 3minutes), and 
when it is running fast, it is instaneous.

One thing I have not done is re-seat the NIC card and/or replace it.  
The symptoms appear to be more hardware than a setup issue.  Especially 
since it goes in and out of the "slow" mode in an unpredictable manner.

Suggestions welcomed.....

Todd

Ross Thomas wrote:

>Scott Doty wrote:
>  
>
>>On Tue, Jul 15, 2003 at 09:39:22AM -0700, Coy Thorp wrote:
>>    
>>
>>>What it sounds like to me, first flush, is a speed/duplex mismatch.  
>>>That means the communication between the network device and the linux 
>>>box is not matched, or the network device and the pc is not matched.  
>>>If one side is 100M, full duplex, then the other side must be that as 
>>>well.  If one side is set to auto-detect the settings, then the other 
>>>side should be set to auto-detect as well.
>>>      
>>>
>>Bingo.  That is my assessment.
>>    
>>
>
>Out of interest, how would this explain the directional asymmetry? 
>Sure, the pipes may be mismatched, but it should still be slow in both 
>directions.  Right?  The only thing that may help a little is buffering 
>in the switch, if any.  But this will still only be an 'initial' 
>speed-up until the buffer fills.
>
>I'm really interested to know how he is testing the throughput.
>
>Ross.
>
>_______________________________________________
>talk mailing list
>talk at nblug.org
>http://nblug.org/mailman/listinfo/talk
>
>.
>
>  
>

-- 

_______________________________________________
talk mailing list
talk at nblug.org
http://nblug.org/mailman/listinfo/talk



More information about the talk mailing list