|
Posted by Al Dykes on April 19, 2005, 10:20 am
If you were Registered and logged in, you could reply and use other advanced thread options
>AnthonyL wrote:
>
>> The arrangement is
>>
>> Office 1 - approx 12 PC's
>> Switch 1 - router- internet
>> |
>> ~ 50 metres
>> |
>> Switch 2
>> Office 2 - approx 30 PC's plus
>> 2 x Windows servers + 1 Novell 3.12 server (IPX)
>>
>> All Cat5e structured cabling.
>>
>> Virtually all traffic is from PC to one or other of the servers in
>> Office 2. There are some networked printers in the system.
>>
>> Traffic load should be relatively low. One central database on SQL
>> server on one server, Word Processing and Exchange on the other and a
>> small DOS application serving 4 users in Office 1 on the Novell
>> server. Some but not heavy internet access via Switch 1.
>>
>> Office 1 staff experience regular delays of sometimes up to a couple
>> of minutes opening files or running an application.
>>
>> The link between the switches is 100mbit (as is the rest of the
>> network except the 10mbit cards in the Novell server).
>>
>> There is a second cable running from Office 1 to Office 2 but that has
>> not been connected and I'm not sure whether it is good practice to
>> have a parallel run.
>>
>> One company has recommended to upgrade the link to a gigabit line but
>> I find the logic hard to follow since the servers will still have
>> 100mbit cards.
>>
>> Any helpful views welcomed.
>
>The question is where the bottleneck lies.
>
>Cheap things to check.
>
>First, run virus, adware, and spyware scans on _everything_ for which you
>have a utility available. If you've got malware of some sort broadcasting
>tons of traffic on your LAN then nothing's going to work right until you
>fix that.
>
> Ping each computer from each other computer and see if there is unusual
>latency in either direction. Download a copy of qcheck
><http://www.ixiacom.com/products/qcheck/>, install it on a Windows box,
>read the readme, and download the appropriate endpoints from the link in
>the readme, then see what speed you're really getting between machines and
>if you're getting the same speed both ways--if you're not then find out
>why--most common cause would be a configuration error--either speed or a
>duplex mismatch.
>
>Windows server includes a tool called "network monitor". Go into the
>Windows Server help and search on that and install it on your servers and
>then watch the traffic going to and from them and see if you're seeing
>anything coming in or going out that doesn't get a response. The free
>version only monitors traffic to and from the server on which it is
>installed, which is a significant limitation, but it's still a useful tool.
>
>There's a tool called "Ethereal" <http://www.ethereal.com/> which is a
>general-purpose network analyzer similar in purpose to "network
>monitor"--ethereal is a free download--install it on one of your machines
>that is experiencing the slowdown and watch the traffic and see if it's
If the OP has two switches then using a traffic monitor will be
impossible unless the switches have management. If the switches have
management then you'll get counters, which might show something
interesting.
Assuming dumb switches I think the blinkenlights are as useful. If
the lights are not blinking he's got a problem with a timeout, not a
high traffic problem. DNS problems would show up as the former. If
you do have lots of blinking lights then the traffic monitor on a
Windows box may tell you something but because you have a switch it
can only see it's own traffic. ISTR that MS Monitor had software that
could be installed at other points in the network and feed data back
to the server for aggregation and analysis. That would solve the switch
limitation.
You can use perfmon.exe on any Windows box (except 98/me/dos) to
analysis in great detail what it's doing.
If your server (CPU and disk as shown by permon) is idle and the
lights are not blinking then you do';t have a capacity problem,
you've got a network design problem. IMO probably DNS.
You can put ethereal on each and every box (except NW) and analysis
the traffic and protocols.
>If you don't already have managed switches (and if you did and knew how to
>use them you'd already know whether the link between them was the
>problem--they'll tell you statistics on it) you might want to consider
>picking up a couple of them off of ebay--Catalyst 2924s go for under $200
>and sometimes under $100. I see Procurve 4000s, which for your use are
>serious overkill but a bit noisy, for about the same.
>
Agreed. Procurve boxes have a lifetime warranty so you can't go
wrong. As for my issue with cheap/noname switch-to-switch
connections, If you match brands, HP or Cisco you should be OK.
--
a d y k e s @ p a n i x . c o m
Don't blame me. I voted for Gore.
| 
NewsGroups 
Yahoo! 
Google 
Windows Live 
del.icio.us 
digg 
Netscape 
XML site map