 |
|
|
|
|
|
| |
| |
|
|
|
|
| |
| |
|
|
OK, So I am still on the hunt for the network delays I am seeing - so
some questions about network throughput.
My Hardware
Windows Domain Server - (1) 1 GB ethernet
File Server - (2) 1GB ethernet (bonded to share load into the same switch)
Patch panel and patch cords are CAT5e
In Wall wiring is combination of CAT5 and CAT5e
Switch is DELL PowerConnect 2724 running in a non-managed mode
about 10 users on the network
What slowdowns might this kind of setup cause?
Thanks
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Le 21/12/2011 15:25, Tom Austin a écrit :
> OK, So I am still on the hunt for the network delays I am seeing - so
> some questions about network throughput.
>
> My Hardware
>
> Windows Domain Server - (1) 1 GB ethernet
>
> File Server - (2) 1GB ethernet (bonded to share load into the same switch)
>
> Patch panel and patch cords are CAT5e
>
> In Wall wiring is combination of CAT5 and CAT5e
>
> Switch is DELL PowerConnect 2724 running in a non-managed mode
>
> about 10 users on the network
>
>
> What slowdowns might this kind of setup cause?
>
How are the lights blinking on the Dell 2724 ?
(if it's a storm or a huge requesting of network resource, some should
blink very fast).
Now, there is other thing that storm in the network to cause a slowdown:
trying to reach/find inaccessible resource and awaiting time-out before
resorting to another solution. In such case the network is rather calm.
(and the light blinks slowly)
In theory Cat5 cable are not suitable for 1G. (cat5e is ok, but not
strict 5)
Have you tested "unloaded network" ? (Keep the WDS & fileserver, as well
as the switch, 1 user only (all other users are disconnected from the
switch). Does it work better ? or same ? or worse ?
Adding 1 user at a time ?
Putting back the Internet connection ?
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 12/21/2011 10:23 AM, Le_Forgeron wrote:
> Le 21/12/2011 15:25, Tom Austin a écrit :
>> OK, So I am still on the hunt for the network delays I am seeing - so
>> some questions about network throughput.
>>
>> My Hardware
>>
>> Windows Domain Server - (1) 1 GB ethernet
>>
>> File Server - (2) 1GB ethernet (bonded to share load into the same switch)
>>
>> Patch panel and patch cords are CAT5e
>>
>> In Wall wiring is combination of CAT5 and CAT5e
>>
>> Switch is DELL PowerConnect 2724 running in a non-managed mode
>>
>> about 10 users on the network
>>
>>
>> What slowdowns might this kind of setup cause?
>>
>
> How are the lights blinking on the Dell 2724 ?
> (if it's a storm or a huge requesting of network resource, some should
> blink very fast).
>
Just making sure - a storm is bad
I'll keep an eye on the switch lights and try to have a look when
someone reports a problem.
Good idea.
> Now, there is other thing that storm in the network to cause a slowdown:
> trying to reach/find inaccessible resource and awaiting time-out before
> resorting to another solution. In such case the network is rather calm.
> (and the light blinks slowly)
>
So, I should label each 'computer' on the switch and look at everything
when there is a problem and pay attention to the 'problem' computer as well.
>
> In theory Cat5 cable are not suitable for 1G. (cat5e is ok, but not
> strict 5)
>
right - that's what I thought. The CAT5 is baked into the building so
it will be hard to replace.
But we can run a test with long patch cables to see if that helps.
> Have you tested "unloaded network" ? (Keep the WDS& fileserver, as well
> as the switch, 1 user only (all other users are disconnected from the
> switch). Does it work better ? or same ? or worse ?
>
We are going to try this next week when the office is pretty empty.
> Adding 1 user at a time ?
> Putting back the Internet connection ?
The main problem is not a small delay in getting a file - 10-15 seconds
can happen from time to time. But 5+ minutes is pretty bad.
Thanks for the tips.
Merry Christmas!
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 12/21/2011 8:45, Tom Austin wrote:
> The main problem is not a small delay in getting a file - 10-15 seconds can
> happen from time to time. But 5+ minutes is pretty bad.
On a gigabit network, that sounds like a long time to load a 5M file.
Do you do a lot of updates on the Linux server, like creating and deleting
thousands of files at a time? It's possible that the Linux server is just
locking up flushing its buffers if it has a lot of RAM. Sadly, I don't know
any sort of monitor that would actually tell you what's going on inside the
kernel of a Linux machine. (Anyone know what the equivalent of Performance
Monitor on Linux is?) If your file server has RAID, maybe there's a bad
spot on one of the disks and it retries a while before falling over to the
other disk? Maybe check through syslog records to see?
Check the domain server's event logs around the time of a problem, and see
if it's an authorization problem or something, where maybe you're waiting
for a new kerberos key to access the remote file and the domain server is
slow giving it out for some reason?
--
Darren New, San Diego CA, USA (PST)
People tell me I am the counter-example.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
Le 21/12/2011 18:00, Darren New nous fit lire :
> On 12/21/2011 8:45, Tom Austin wrote:
>> The main problem is not a small delay in getting a file - 10-15
>> seconds can
>> happen from time to time. But 5+ minutes is pretty bad.
>
> On a gigabit network, that sounds like a long time to load a 5M file.
On gigabit network, a 5M file which should transfer in about 4000
frames, should take less than 10s even at 1% of network usage, including
an acknowledge of each frame.
(1% of network: which means that even on a 10mbits, 10s is already large)
>
> Do you do a lot of updates on the Linux server, like creating and
> deleting thousands of files at a time? It's possible that the Linux
> server is just locking up flushing its buffers if it has a lot of RAM.
> Sadly, I don't know any sort of monitor that would actually tell you
> what's going on inside the kernel of a Linux machine. (Anyone know what
> the equivalent of Performance Monitor on Linux is?)
there is:
* text mode: top
* gnome : gnome-system-monitor
* kde : ksysguard
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 12/21/2011 12:00 PM, Darren New wrote:
> On 12/21/2011 8:45, Tom Austin wrote:
>> The main problem is not a small delay in getting a file - 10-15
>> seconds can
>> happen from time to time. But 5+ minutes is pretty bad.
>
> On a gigabit network, that sounds like a long time to load a 5M file.
>
Agreed, but the occasional 'glitch' of 10-15 seconds is acceptable.
> Do you do a lot of updates on the Linux server, like creating and
> deleting thousands of files at a time? It's possible that the Linux
> server is just locking up flushing its buffers if it has a lot of RAM.
> Sadly, I don't know any sort of monitor that would actually tell you
> what's going on inside the kernel of a Linux machine. (Anyone know what
> the equivalent of Performance Monitor on Linux is?) If your file server
> has RAID, maybe there's a bad spot on one of the disks and it retries a
> while before falling over to the other disk? Maybe check through syslog
> records to see?
>
No updates, but a lot of large (5 MB) image files are being transferred
to and around the file system.
It is 32 bit - so only 4GB of physical ram with some extra swap just in
case.
It does have RAID - OS is RAID 1 on the software level, file drives are
RAID 1 on a 3Ware controller 2 mirrored sets as a logical volume.
> Check the domain server's event logs around the time of a problem, and
> see if it's an authorization problem or something, where maybe you're
> waiting for a new kerberos key to access the remote file and the domain
> server is slow giving it out for some reason?
>
That could be a very real possibility - and would bring our crappy
Windows domain server into question - like we want it to be :-)
This is good food for thought. I'll set up some tests that might help
flush some of this out.
Thanks
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
On 12/21/2011 12:38 PM, Le_Forgeron wrote:
>
> On gigabit network, a 5M file which should transfer in about 4000
> frames, should take less than 10s even at 1% of network usage, including
> an acknowledge of each frame.
> (1% of network: which means that even on a 10mbits, 10s is already large)
>
yes, agreed - but an occasional hiccup is tolerable so we didn't worry
about it. Now with more users and data maybe that hiccup 'error' is
being compounded.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> OK, So I am still on the hunt for the network delays I am seeing - so
> some questions about network throughput.
>
> My Hardware
>
> Windows Domain Server - (1) 1 GB ethernet
>
Make sure the server NICs are hard-coded for 1 Gbps Full-duplex.
Gigabitethernet left in auto-detect mode can lead to funky things...
Make sure the switch ports facing the servers also have their speed and
duplex hardcoded, if possible.
> File Server - (2) 1GB ethernet (bonded to share load into the same switch)
>
How? Depending on the way it's set up, it can cause major problems with
the switch. Disconnect one of the patch cords and see if performance
improves. If it does, you had a loop between the two switch ports, and
the two NICs in your server. Either reconfigure your NICs to use
fail-over, or make sure "spanning-tree" is enabled on the switch ports
facing the server. Keep in mind that by doing that you will lose your
2Gbps aggregation speed and the second NIC will only be there to provide
a host-stanby.
> Patch panel and patch cords are CAT5e
>
Length? CAT5e specification says that desks-side patch cords must not
be longer than 3m (10ft), and that switch-side patch cords can not be
longer than 7m (25ft), while the total length of the premises cabling
(in the wall) must not exceed 90m (300ft). See below to know how to do
this.
> In Wall wiring is combination of CAT5 and CAT5e
>
Do you have lots of internediary patch panel connections (e.g.: desk to
wiring closet. Wiring closet to server room main wall. Server room
wall to rack patch panel)
> Switch is DELL PowerConnect 2724 running in a non-managed mode
>
According to TFM (which you should R), there are cable-testing functions
in the switch. Check to see if you have faulty cabling, exceed the
length of the standards, or if the premises cabling can really sustain
1Gbps. If not, downgrade the user ports to 100 Mbps.
> about 10 users on the network
>
Check, if possible, for rogue user activity (Is anyone running a
streaming app? Is anyone acting as a Counterstrike server? Does anyone
have NetBEUI turned on by mistake? etc...)
>
> What slowdowns might this kind of setup cause?
>
Apart from the risk of having a spanning-tree loop between the switch
and the dual-homed server, there's nothing inherently bad about this
setup. By default it should run fine.
>
> Thanks
>
>
>
>
--
/*Francois Labreque*/#local a=x+y;#local b=x+a;#local c=a+b;#macro P(F//
/* flabreque */L)polygon{5,F,F+z,L+z,L,F pigment{rgb 9}}#end union
/* @ */{P(0,a)P(a,b)P(b,c)P(2*a,2*b)P(2*b,b+c)P(b+c,<2,3>)
/* gmail.com */}camera{orthographic location<6,1.25,-6>look_at a }
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
>
> How? Depending on the way it's set up, it can cause major problems with
> the switch. Disconnect one of the patch cords and see if performance
> improves. If it does, you had a loop between the two switch ports, and
> the two NICs in your server. Either reconfigure your NICs to use
> fail-over, or make sure "spanning-tree" is enabled on the switch ports
> facing the server. Keep in mind that by doing that you will lose your
> 2Gbps aggregation speed and the second NIC will only be there to provide
> a host-stanby.
>
And by that, I mean "hot-standby". I Should have proof-read...
--
/*Francois Labreque*/#local a=x+y;#local b=x+a;#local c=a+b;#macro P(F//
/* flabreque */L)polygon{5,F,F+z,L+z,L,F pigment{rgb 9}}#end union
/* @ */{P(0,a)P(a,b)P(b,c)P(2*a,2*b)P(2*b,b+c)P(b+c,<2,3>)
/* gmail.com */}camera{orthographic location<6,1.25,-6>look_at a }
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
| |
|
|
> OK, So I am still on the hunt for the network delays I am seeing - so
> some questions about network throughput.
>
> My Hardware
>
> Windows Domain Server - (1) 1 GB ethernet
>
> File Server - (2) 1GB ethernet (bonded to share load into the same switch)
>
> Patch panel and patch cords are CAT5e
>
> In Wall wiring is combination of CAT5 and CAT5e
>
> Switch is DELL PowerConnect 2724 running in a non-managed mode
>
> about 10 users on the network
>
>
> What slowdowns might this kind of setup cause?
>
>
> Thanks
>
>
>
>
I know someone who had slow network. He discovered that he was
configured as peer to peer instead of on a domain. This can realy slow
things down.
Check for crushed or otherwise damaged cables. Any work on a wall
containint cabling? It could have damaged some cabling.
I think that, at least sometimes, a faulty network adaptor on one
computer may cause erratic trafic over the network.
Post a reply to this message
|
|
| |
| |
|
|
|
|
| |
|
|
