In the off chance someone has an idea, what would be a logical 
explanation to a couple of windows machines, linked via a simple cheap 
switch, both forced to full duplex 100mb - taking nearly 2 days to 
transfer 10 2gb files from one box to the other?

Should I focus on the core operating system, NIC's or switch. I have my 
own inclination - but some other takes on it would be most welcome.

-- 
. . .

Stephen Ward wrote:
> In the off chance someone has an idea, what would be a logical 
> explanation to a couple of windows machines, linked via a simple cheap 
> switch, both forced to full duplex 100mb - taking nearly 2 days to 
> transfer 10 2gb files from one box to the other?
> 
> Should I focus on the core operating system, NIC's or switch. I have my 
> own inclination - but some other takes on it would be most welcome.
> 

That sounds a bit long .... for a 10Mb hub :-(

A bit more info could be useful: which operating system and service pack
(ISTR Vista had a problem with indexing and transfers for a large number
of files when the receiving folder was open);

why "force" 100Mb? what happens if you use auto-negotiate?

have you tried different cables?

-- 
PeeGee

"Nothing should be able to load itself onto a computer without the
knowledge or consent of the computer user. Software should also be able
to be removed from a computer easily."
Peter Cullen, Microsoft Chief Privacy Strategist (Computing 18 Aug 05)

On Sun, 04 Jan 2009 10:26:15 +0000, PeeGee wrote:

> Stephen Ward wrote:
>> In the off chance someone has an idea, what would be a logical
>> explanation to a couple of windows machines, linked via a simple cheap
>> switch, both forced to full duplex 100mb - taking nearly 2 days to
>> transfer 10 2gb files from one box to the other?
>> 
>> Should I focus on the core operating system, NIC's or switch. I have my
>> own inclination - but some other takes on it would be most welcome.
>> 
>> 
> That sounds a bit long .... for a 10Mb hub :-(
> 
> A bit more info could be useful: which operating system and service pack
> (ISTR Vista had a problem with indexing and transfers for a large number
> of files when the receiving folder was open);
> 
> why "force" 100Mb? what happens if you use auto-negotiate?
> 
> have you tried different cables?

There are a pair of old XP SP2 machines. I forced 100mbps FD on both in 
an attempt to speed things up. It's made no difference from the previous 
'auto-neg'. I don't get any issues when I kick stuff through the same 
switch with smb on my linux boxes - it's just these two 'laughing boys' 
that are acting up.

Cables have been changed (yep - been bitten on that before), switch ports 
etc.



-- 
. . .

In article <49607eca$0$27148$fa0fcedb@news.zen.co.uk>, Stephen Ward 
says...
> In the off chance someone has an idea, what would be a logical 
> explanation to a couple of windows machines, linked via a simple cheap 
> switch, both forced to full duplex 100mb - taking nearly 2 days to 
> transfer 10 2gb files from one box to the other?
> 
> Should I focus on the core operating system, NIC's or switch. I have my 
> own inclination - but some other takes on it would be most welcome.
> 
Disable your AV software.

-- 
Conor

I only please one person per day. Today is not your day. Tomorrow isn't 
looking good either. - Scott Adams

Stephen Ward wrote:
> In the off chance someone has an idea, what would be a logical 
> explanation to a couple of windows machines, linked via a simple cheap 
> switch, both forced to full duplex 100mb - taking nearly 2 days to 
> transfer 10 2gb files from one box to the other?

Duplex mismatch seems a likely explanation. This results in erratic 
transfer rates (see Task Manager Networking tab), usually no more than 
200KB/s average, many Ethernet-level errors and TCP segment 
retransmissions (see Network Interface and TCP performance counters in 
perfmon).

Try forcing one or both ends to half duplex. For what it's worth, full 
duplex makes little difference to performance if data is primarily going 
one way - on an in-spec network, very little bandwidth is lost in 
collisions.

Alex

On Sun, 04 Jan 2009 12:45:39 +0000, Conor wrote:

> In article <49607eca$0$27148$fa0fcedb@news.zen.co.uk>, Stephen Ward
> says...
>> In the off chance someone has an idea, what would be a logical
>> explanation to a couple of windows machines, linked via a simple cheap
>> switch, both forced to full duplex 100mb - taking nearly 2 days to
>> transfer 10 2gb files from one box to the other?
>> 
>> Should I focus on the core operating system, NIC's or switch. I have my
>> own inclination - but some other takes on it would be most welcome.
>> 
> Disable your AV software.

I'd already done that but thanks for the suggestion. Also kicked out the 
old free 'zoneminder' that was running just in case that was messing with 
it, but no dice.

I've got to thank Alex for his suggestions - my networking tab on task 
manager gives me no throughput speed. It just shows the link at 100m and 
a couple of % utilisation.



-- 
. . .

Alex Fraser wrote:
> Stephen Ward wrote:
>> In the off chance someone has an idea, what would be a logical 
>> explanation to a couple of windows machines, linked via a simple cheap 
>> switch, both forced to full duplex 100mb - taking nearly 2 days to 
>> transfer 10 2gb files from one box to the other?
> 
> Duplex mismatch seems a likely explanation. This results in erratic 
> transfer rates (see Task Manager Networking tab), usually no more than 
> 200KB/s average, many Ethernet-level errors and TCP segment 
> retransmissions (see Network Interface and TCP performance counters in 
> perfmon).
> 
> Try forcing one or both ends to half duplex. For what it's worth, full 
> duplex makes little difference to performance if data is primarily going 
> one way - on an in-spec network, very little bandwidth is lost in 
> collisions.
> 
> Alex
I've had problems with XP when trying to do more than 1 copy/move at the 
same time between mapped drives. Everything grinds to a halt, I now try 
to be patient and run one copy/move window at a time.

On Sun, 04 Jan 2009 17:10:50 +0000, robert wrote:

> Alex Fraser wrote:
>> Stephen Ward wrote:
>>> In the off chance someone has an idea, what would be a logical
>>> explanation to a couple of windows machines, linked via a simple cheap
>>> switch, both forced to full duplex 100mb - taking nearly 2 days to
>>> transfer 10 2gb files from one box to the other?
>> 
>> Duplex mismatch seems a likely explanation. This results in erratic
>> transfer rates (see Task Manager Networking tab), usually no more than
>> 200KB/s average, many Ethernet-level errors and TCP segment
>> retransmissions (see Network Interface and TCP performance counters in
>> perfmon).
>> 
>> Try forcing one or both ends to half duplex. For what it's worth, full
>> duplex makes little difference to performance if data is primarily
>> going one way - on an in-spec network, very little bandwidth is lost in
>> collisions.
>> 
>> Alex
> I've had problems with XP when trying to do more than 1 copy/move at the
> same time between mapped drives. Everything grinds to a halt, I now try
> to be patient and run one copy/move window at a time.

It finally finished so to try and trouble shoot I've copied one 2g file 
from each box through the same switch to a samba share on a linux box. 
This was much quicker. It's really odd behaviour to see. The issue seems 
to be when the only xp boxes I have try to talk. I may look at spending 
some more time with it.



-- 
. . .

Stephen Ward wrote:
[snip]
> It finally finished so to try and trouble shoot I've copied one 2g file 
> from each box through the same switch to a samba share on a linux box. 
> This was much quicker. It's really odd behaviour to see. The issue seems 
> to be when the only xp boxes I have try to talk. I may look at spending 
> some more time with it.

Do you see high CPU usage on either end? (On a relatively modern 
machine, 100Mbit transfer should barely register.)

I've seen some strange behaviour on occasions caused by different 
Windows versions, or different applied updates, that caused problems; 
one obvious effect was that browsing a share was slow, with a 
significant added delay for every level in the directory tree. I can't 
remember if this affected transfer performance.

I don't have any particular recommendations for software (hopefully 
someone else will suggest something), but it might be useful to try some 
TCP throughput tests between the machines - if these are fine then it 
strongly suggests the problem is specific to file sharing rather than a 
more fundamental network problem.

Alex

On 04 Jan 2009 18:00:01 GMT, Stephen Ward
 wrote:

>On Sun, 04 Jan 2009 17:10:50 +0000, robert wrote:
>
>> Alex Fraser wrote:
>>> Stephen Ward wrote:
>>>> In the off chance someone has an idea, what would be a logical
>>>> explanation to a couple of windows machines, linked via a simple cheap
>>>> switch, both forced to full duplex 100mb - taking nearly 2 days to
>>>> transfer 10 2gb files from one box to the other?
>>> 
>>> Duplex mismatch seems a likely explanation. This results in erratic
>>> transfer rates (see Task Manager Networking tab), usually no more than
>>> 200KB/s average, many Ethernet-level errors and TCP segment
>>> retransmissions (see Network Interface and TCP performance counters in
>>> perfmon).
>>> 
>>> Try forcing one or both ends to half duplex. For what it's worth, full
>>> duplex makes little difference to performance if data is primarily
>>> going one way - on an in-spec network, very little bandwidth is lost in
>>> collisions.
>>> 
>>> Alex
>> I've had problems with XP when trying to do more than 1 copy/move at the
>> same time between mapped drives. Everything grinds to a halt, I now try
>> to be patient and run one copy/move window at a time.
>
>It finally finished so to try and trouble shoot I've copied one 2g file 
>from each box through the same switch to a samba share on a linux box. 
>This was much quicker. It's really odd behaviour to see. The issue seems 
>to be when the only xp boxes I have try to talk. I may look at spending 
>some more time with it.

try wireshark on 1 or both of the PCs.

you should be able to capture part of the transfer and see what isnt
happening :)
-- 
Regards

stephen_hope@xyzworld.com - replace xyz with ntl

On Sun, 04 Jan 2009 21:35:11 +0000, Stephen wrote:

> On 04 Jan 2009 18:00:01 GMT, Stephen Ward
>  wrote:
> 
>>On Sun, 04 Jan 2009 17:10:50 +0000, robert wrote:
>>
>>> Alex Fraser wrote:
>>>> Stephen Ward wrote:
>>>>> In the off chance someone has an idea, what would be a logical
>>>>> explanation to a couple of windows machines, linked via a simple
>>>>> cheap switch, both forced to full duplex 100mb - taking nearly 2
>>>>> days to transfer 10 2gb files from one box to the other?
>>>> 
>>>> Duplex mismatch seems a likely explanation. This results in erratic
>>>> transfer rates (see Task Manager Networking tab), usually no more
>>>> than 200KB/s average, many Ethernet-level errors and TCP segment
>>>> retransmissions (see Network Interface and TCP performance counters
>>>> in perfmon).
>>>> 
>>>> Try forcing one or both ends to half duplex. For what it's worth,
>>>> full duplex makes little difference to performance if data is
>>>> primarily going one way - on an in-spec network, very little
>>>> bandwidth is lost in collisions.
>>>> 
>>>> Alex
>>> I've had problems with XP when trying to do more than 1 copy/move at
>>> the same time between mapped drives. Everything grinds to a halt, I
>>> now try to be patient and run one copy/move window at a time.
>>
>>It finally finished so to try and trouble shoot I've copied one 2g file
>>from each box through the same switch to a samba share on a linux box.
>>This was much quicker. It's really odd behaviour to see. The issue seems
>>to be when the only xp boxes I have try to talk. I may look at spending
>>some more time with it.
> 
> try wireshark on 1 or both of the PCs.
> 
> you should be able to capture part of the transfer and see what isnt
> happening :)

I already tried a tcpdump and looked at it in wireshark. The TCP 
conversation is normal (SYN/ACK) - no retransmissions or anything odd 
going on. I'll look at it a bit more closely in a couple of weeks. I 
don't use Windows that much for it to be an issue, I'd just like to get 
to the bottom of it. Thanks for everybody who took the time and trouble 
to make suggestions. Much appreciated.



-- 
. . .