Re: Network speed
- From: "Nige" <nigelDOTINCH@xxxxxxxxxxxxxxxxxxxxxxx>
- Date: Sun, 15 Mar 2009 14:49:51 -0000
bod43 wrote:
On 15 Mar, 12:08, Paul Carmichael <wibbleypa...@xxxxxxxxx> wrote:
On Sun, 15 Mar 2009 11:49:02 +0000, TMack escribió:
Paul Carmichael wrote:
Anybody here have an idea why my LAN should be moving data at a
miserable
4.5MB/s over the cable? The router's new.
Depends what you are transferring and from where. If its lots of
smallish files from a badly fragmented hard disk then the overall
transfer rate from the disk comes into play. The sustained transfer
rate of a drive is not the only consideration. If you had 196,916 2KB
files, the transfer time (when the head gets there), would be virtually
zero for each file, but it takes milliseconds to find each file ( the
disk "seeking" the heads to the appropriate track). If you can process a
maximum of 500 of the 2KB files each second, due to the limitations of
head movement, that is 1MB/sec transfer rate, and looks pathetic. If the
files were fragmented then it would take even longer for each one to
transfer, due to additional time taken for the heads to find and
transfer each fragment.
It's coming from a sata2 ntfs disk and it's a single 9GB file. The xp
installation is pretty new. I'm beginning to suspect it's an OS
bottleneck.
Just check that your PCs are set to Speed_Auto/Duplex_Auto.
By definition one end of a wire on Auto_Dup and the other on
fixed_dup is *broken*. Doubt that your speedtouch is configurable
away from Auto so the PCs will *have* to be auto too.
There should be *zero* (or close to, 10E-8 is OK) errors
reported with "netstat -e".
Test the *network* with ttcp/iperf/choose_your_poison.
Run multiple instances (say up to 20/30/50) until the total throughput
stops increasing. this factors aout any network stack buffering
issues.
iperf -P # if i recall correctly does multiple
# streams over one session.
Defrag the disks. On Windows run it more than once.
Windows SMB (CIFS, Windosw Shares, choose your name)
uses 64k blocks at most (in practise inevitably 32k)
so this I suppose *might* be limiting the transfer but
if the computers are on the same LAN I wouldn't have
thought so.
Worth trying xcopy since some windows explorer
copying behaved (now fixed) badly in one direction
or the other due to being optimised to keep the file
list window fully and exactly and completely up to date
in preference to doing copying.
Worth considering ftp but in practise they are all broken too
with small hidden internal buffers.
BTW the recent posted wireless throughpout is a red herring.
54Mbps wireless can only transfer at about half that by
design. I am not sure of the details. 100M Ethernet can transfer
at 100M by design - I am sure of those details. The only overhead
is the trivial IP/TCP encapsulation overhead of about say 60/1500.
What did he say?
--
Nige,
Range Rover Td6 Vogue
BMW K1200S
Honda VTR1000F
Discovery II Td5 Auto
.
- References:
- OT: Network speed
- From: Paul Carmichael
- Re: Network speed
- From: TMack
- Re: Network speed
- From: Paul Carmichael
- Re: Network speed
- From: bod43
- OT: Network speed
- Prev by Date: Re: OT: Wanted - software to rotate JPG
- Next by Date: Re: OT: Open letter to Sir Liam Donaldson
- Previous by thread: Re: Network speed
- Next by thread: Re: Network speed
- Index(es):
Relevant Pages
|
