Is my 100 megabit network now a bottleneck ? Should I upgrade to 1GB ?
John...@inbox.ru
another on the same machine. It takes 8 minutes to copy the same file
from one machine to another over the 100 megabit network (source and
destination drives are very similar hardware wise, both 160 GB Seagate,
one SATA, another - ATA 100, same generation), and the network
utilization is about 85-90%.
Does it mean the network is a bottleneck ? Will I gain anyting by
upgrading to 1 gigabit considering that prices on this equipment are
now reasonable ? What is the importance of Gigabit switch supporting
Jumbo frames ?
Walter Roberson
<John...@inbox.ru> wrote:
:It takes 1 min 25 seconds to copy 4GB file from one physical drive to
:another on the same machine. It takes 8 minutes to copy the same file
:from one machine to another over the 100 megabit network (source and
:destination drives are very similar hardware wise, both 160 GB Seagate,
:one SATA, another - ATA 100, same generation), and the network
:utilization is about 85-90%.
:Does it mean the network is a bottleneck ?
That's about 71.5 megabits/s, which is not terrible for tcp but could
be better.
:Will I gain anyting by
:upgrading to 1 gigabit considering that prices on this equipment are
:now reasonable ?
I suggest that you test your network using ttcp . That will give you
an idea of whether the bottleneck is your disk or your network.
You don't happen to mention your operating system or NIC or anything
about the architecture of your systems. You also don't happen to mention
what tool you are using to do the copy. We don't really have enough
information to tell from the given inforrmation where the bottleneck
is.
With that data rate, my -suspicion- is that you are starting to hit the
limit either of your NIC or of your architecture transfering data to or
from the NIC. If it is your NIC, then a different NIC might get you
a bit better performance even with everything else the same.
If it's the architecture, then going gig won't necessarily help at all.
:What is the importance of Gigabit switch supporting
:Jumbo frames ?
That depends greatly on your traffic patterns. Do you have a managed
switch now? If so then look at the packet size histograms. If, as is
common, most of your packets are < 256 bytes, then jumbo frames
aren't going to help very much. If a fair portion of your packets
are of maximum size, then jumbo frames could be beneficial.
Jumbo frames are of greatest use when you are often transfering large
files. They reduce the overhead of IP transmissions by reducing the
number of times the MACs and IPs and TCP sequence numbers and so on
need to be transmitted -- by about 60*6*8 bit-times per jumbo frame
compared to regular sized frames. That's not very much at all at
gigabit speeds, so you have to look at other factors if you are
transmitting on a local LAN. Consider the advantages when you have a
large latency (many fewer ACKs). Consider too the effects when
you have noticable contention for a shared link --- getting 6 times
as much data through per instance that you are able to seize the
link can be a significant advantage if other devices are usually
waiting to transmit...
--
Look out, there are llamas!
williamp.n.smith
>It takes 1 min 25 seconds to copy 4GB file from one physical drive to
>another on the same machine. It takes 8 minutes to copy the same file
>from one machine to another over the 100 megabit network (source and
>destination drives are very similar hardware wise, both 160 GB Seagate,
>one SATA, another - ATA 100, same generation), and the network
>utilization is about 85-90%.
I'm getting about 200 megabits utilization on my gigabit network
between two relatively modern machines (P4-2.4, SATA drives,
integrated Intel gigabit controllers on the FSB, etc.), so you could
probably expect to gain something.
John...@inbox.ru
> I suggest that you test your network using ttcp . That will give you
> an idea of whether the bottleneck is your disk or your network.
I will do this.
> You don't happen to mention your operating system or NIC or anything
> about the architecture of your systems. You also don't happen to
mention
> what tool you are using to do the copy. We don't really have enough
> information to tell from the given inforrmation where the bottleneck
> is.
One computer is 3.6 GHZ P4 with 1 GB of RAM and onboard Broadcom
Gigabit network adapter, another - 1 GHZ PIII with Linksys PCI 100
mebabit network card. They are connected through Linksys BEFSR41
configured as a switch.
Is there some kind of test progrmam to find out if onboard Broadcom
Gigabit network adapter supports Jumbo Frames ? I couldn't find any
info on that on Dell website. I submitted support request but haven't
received the answer yet.
Walter Roberson
<John...@inbox.ru> wrote:
:One computer is 3.6 GHZ P4 with 1 GB of RAM and onboard Broadcom
:Gigabit network adapter, another - 1 GHZ PIII with Linksys PCI 100
:mebabit network card. They are connected through Linksys BEFSR41
:configured as a switch.
I would look first at the BEFSR41 as being the bottleneck.
The BEFSR41 is not designed as a high performance switch.
:Is there some kind of test progrmam to find out if onboard Broadcom
:Gigabit network adapter supports Jumbo Frames ?
http://darkwing.uoregon.edu/~joe/jumbo-clean-gear.html
9000 MTU Ethernet
Martin Kuhne mentioned "Broadcom BCM5703 (64-bit) (3Com 3C996B-T
($65), Dell PowerEdge integrated BCM570x NIC, Compaq NC7770) 9k,
verified working"
--
Disobey all self-referential sentences!
John...@inbox.ru
PCAUSA Test TCP Utility V2.01.01.07
TCP Receive Test
Local Host : xpsr1000
**************
Listening...: On port 5001
Accept : TCP <- 192.168.3.80:1117
Buffer Size : 8192; Alignment: 16384/0
Receive Mode: Sinking (discarding) Data
Statistics : TCP <- 192.168.3.80:1117
16777216 bytes in 1.41 real seconds = 11603.40 KB/sec +++
numCalls: 2057; msec/call: 0.70; calls/sec: 1456.80
AFAIU this is close to theoretical limit for 100Mbit network. So it
seems that Linksys router is not a bottleneck. Right ?
Regarding network card supporting - I found setting to turn on Jumbo
Frames on the Advanced Tab on my work machine (Dell SC400 server). My
home machine (the one this post is about) doesn't have this option on
the Advanced Tab, most likely it doesn't support it (Dell tech support
say it does, but where am I supposed to turn it on then?).
John...@inbox.ru
PCAUSA Test TCP Utility V2.01.01.07
TCP Receive Test
Local Host : xpsr1000
**************
Listening...: On port 5001
Accept : TCP <- 192.168.3.80:1117
Buffer Size : 8192; Alignment: 16384/0
Receive Mode: Sinking (discarding) Data
Statistics : TCP <- 192.168.3.80:1117
16777216 bytes in 1.41 real seconds = 11603.40 KB/sec +++
numCalls: 2057; msec/call: 0.70; calls/sec: 1456.80
AFAIU this is close to theoretical limit for 100Mbit network. So it
seems that Linksys router is not a bottleneck. Right ?
Regarding network card supporting - I found setting to turn on Jumbo
Frames on the Advanced Tab on my work machine (Dell SC400 server). My
home machine (Dell Dimension 8400, the one this post is about) doesn't
Walter Roberson
<John...@inbox.ru> wrote:
:OK so here is my TTCP results:
:16777216 bytes in 1.41 real seconds = 11603.40 KB/sec +++
:AFAIU this is close to theoretical limit for 100Mbit network. So it
:seems that Linksys router is not a bottleneck. Right ?
Correct, that is near the limit. I would suggest running the
test for longer (e.g., increase the number of buffers) so as to
reduce round-off effects and start-up effects.
You can get a direct readout in megabits/s with one of the ttcp
options; on unix it would be -f m .
As a point of comparison, I get 90-91 Mbit/s between two non-Intel
server systems with Tigon 3 cards in them, that are being limited
by the 100 Mbit switch they are connected to.
--
I don't know if there's destiny,
but there's a decision! -- Wim Wenders (WoD)
anoop
John...@inbox.ru wrote:
[..]
> What is the importance of Gigabit switch supporting
> Jumbo frames ?
This is mostly of importance in reducing the per-packet
processing at the endpoints, so if one or both endpoints
is the bottleneck, jumbos would definitely help.
Anoop
williamp.n.smith
>One computer is 3.6 GHZ P4 with 1 GB of RAM and onboard Broadcom
>Gigabit network adapter, another - 1 GHZ PIII with Linksys PCI 100
>mebabit network card. They are connected through Linksys BEFSR41
>configured as a switch.
The first one will definately handle over 100 megabits (I have one
machine in that class, and it does ~200 megabits doing Windows file
copies). The second one is more dicey. You could always spend the
$100 on a new card for the second machine and a gigabit switch and see
what happens...
ErwinWieser
month ago and it works much better than the one with 100mbpersecond!!!!
greetings |blackhawk.ATS| (Return to Castle Wolfenstein: Enemy Territory)
<John...@inbox.ru> schrieb im Newsbeitrag
news:1103561263.5...@f14g2000cwb.googlegroups.com...