Tek-Tips is the largest IT community on the Internet today!

Members share and learn making Tek-Tips Forums the best source of peer-reviewed technical information on the Internet!

  • Congratulations gkittelson on being selected by the Tek-Tips community for having the most helpful posts in the forums last week. Way to Go!

Gigabit network is very slow..try to improve performance

Status
Not open for further replies.

mwilliamsess

IS-IT--Management
Dec 29, 2005
12
US
I have been working with this company for a few months now and I am trying to improve the network performance. The person who setup this network no longer works here. We are using 3 different subnets on our LAN. They are private 10.x.x.x with mask 255.255.255.0. We have about 20 servers and each is assigned a static ip from each subnet. Behind our router and firewall we have a layer 3 switch that has the routes for each of the subnets. Our environment consists of about 500 computers, 30 switchs, and probably over 50 networked printers. It seems odd that we are using 3 different subnets that have to be routed internally through the layer 3 switch. Our company has been assigned a block of address to use in the 10.x.x.x range. Would we benefit by taking the address block we have and using CIDR? This would allow us to use 1 large subnet and do away with the internal routing.
 
If you put 500 hosts on the same subnet you will greatly reduce the efficiency of your network. If all machines are in the same broadcast domain, then every broadcast packet must be sent to every host. Not a very efficient way to set up your network.

But that doesn't mean that the subnetting that has been done has necessarily been accomplished efficiently. I would guess, though, by your description, that a bit of thought was put into the architecture.

Generally you will subnet geographically, like by floor, or by building, maintaining the majority of your traffic in a single geographical location. That way, the only traffic that is routed is traffic that needs to go across a WAN or Backbone to get to another geographic area.

Don't think of internal routing as a bad thing. It generally only adds a small amount of delay (in the neighborhood of 7ms). And because every machine in the network is not having to filter ARP gateway requests for all 500 machines, each machine is doing less work.

BTW, you were not "assigned" a 10.x.x.x address block. That address block is part of the RFC 1918 Private Network Address pool that anyone can use freely. It cannot be routed across the Internet.

With an address block of that size, using a Class C netmask is not uncommon. If you are running out of address space (you aren't), or creating small WAN or transmission networks, it is not uncommon to migrate to addressing that is completely classless, like CIDR. You could specify your current netmasks as CIDR addresses by using /24 if you really want to go that route. CIDR doesn't really buy you anything, it is just a newer netmask convention that specifies the number of bits in the network portion of the address. For every CIDR network size, there is an equivalent netmask. A /30 CIDR address is the same as a 255.255.255.252. A /29 CIDR address is the same as a 255.255.255.248.


pansophic
 
I don't know the reason for having 3 IP-networks in your installation, but at our installation we have around 100 servers and 300 workstations and a bunch of printers on the inside.
The config we have (I did the "design") is based on a very simple layer 2 setup.
We have 7 floors each connected via 2 x 1Gbit/s connections and all servers (at least almost) are running 1Gbit/s.
Broadcast has never been a problem and we have an external company to check our lan 4 times a year.

This setup has been in production for 5 years and we have never seen any kind of broadcast problems.

You core switch has support for layer 3 switching, so the design we have was done for easy use and maintenance and performance. Layer 2 switching is a lot faster then layer 3 switching for a network this size (without any WAN connections)

The hardest load on our LAN I have ever seen was something like 20 Gbit/s total switching at a peak (don't remember the PPS)

/johnny
 
What are some good tools I can use to monitor the broadcast on our network. Our network just doesn't run fast at all. I noticed today that our switches are almost 2 years behind on software updates. I am trying to figure out where are problem is. I was planning on isolating different legs of our network and doing some file copies for speed test.
 
You will need a protocol analyzer to check into it.

Ethereal is a nice freeware tool (
HP (Agilent) has some nice protocol analyzers. But they cost a lot.

Configure a mirror port on the switch you want to monitor and take a normal PC with Ethereal and capture your traffic for 5-10 min and then start the hard work to analyze the traffic you see. (This req. that you know how to analyze your network protocols)

But do you know for sure that your problem is broadcast?
It could be as simple as some bad cables and/or NIC's.

Another common problem is that some NIC's run 100 Mbit/s and autosense. That can also kill performance.

/johnny
 
Status
Not open for further replies.

Part and Inventory Search

Sponsor

Back
Top