i thought i read somewhere that when you go to a switched infrastructure that putting a gig card in the file server will increase performance on the network. I think the logic was that with all the computers talking at 100 on their own segment, the server can be a bottleneck. However, i'm wondering if that is technically accurate. Do i understand correctly that even with a switch, only one node is communicating with a server at one time? If so, wouldn't the 'rate determining step' be the workstation?
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 Chriss Miller on being selected by the Tek-Tips community for having the most helpful posts in the forums last week. Way to Go!
will gig nic in file server improve thru-put in switched lan
- Thread starter johnpet
- Start date
davemiller
Technical User
the chain is only as strong as the weakest link.
you need to know the design of your lan and wan, and the demands on it in order to really understand how you can improve throughput.
generally speaking the gig addition to your server should
provide some improvement.
you need to know the design of your lan and wan, and the demands on it in order to really understand how you can improve throughput.
generally speaking the gig addition to your server should
provide some improvement.
It all depends on your appllications and how they connect with the server. You can have all the bandwidth in the world, if your sever can't handle it then whats the point.
From a pure networking standpoint then yes it will improve your LAN performance. If you have a test set say an Adtech or Ixia and you run pure IP traffic from 10 100M ports to one Gig port you will see all the packets get through. If you run 10 100Mports to 1 100M port you will see alot of drops.
However, you have TCP/IP slow start and other factors that contirbute to the overall throughput of the network.
So, I'm not sure what you mean by saying only one node talks to a server at a time. Most servers Today are multi threaded and can support serveral concurrent sessions.
I hope this helps.
From a pure networking standpoint then yes it will improve your LAN performance. If you have a test set say an Adtech or Ixia and you run pure IP traffic from 10 100M ports to one Gig port you will see all the packets get through. If you run 10 100Mports to 1 100M port you will see alot of drops.
However, you have TCP/IP slow start and other factors that contirbute to the overall throughput of the network.
So, I'm not sure what you mean by saying only one node talks to a server at a time. Most servers Today are multi threaded and can support serveral concurrent sessions.
I hope this helps.
First yes only machine is talking at a time, but that's not a concern at these rates.
Second, can your cable plant handle Gig traffic? If not the card will (defaultly installed) auto-negotiate the next fastest connection, say 100 MB. I assume you already have that in current hardware. Below will discuss optimizing your netowrk and/or replacing components.
How many users do you have? Keep it simple, there may be no need to upgrade. These are the questions that you have to answer before you can even guess at your network's performance.
Are all of the clients on a hub or dedicated to a switch port? If they are dedicated to a switchport, then cut the client network cards to auto-negotiate and throttle the switch ports back half duplex. This will force each client workstation to 10MB. then put the server in the switch dedicated at 100 MB. This will keep any one machine from momopolizing the Server's resources. All users will see an improvement, unless you have 150 and one server.
If you can take the server and get it out of the broadcast domain with the clients, use helper-addresses on a router if it's also a dhcp server. This will reduce the amount of broadcast traffic the server has to process. (Not sure that an application server makes a good DHCP server, but hey you know your budget and your network size.
Take metrics of your current bandwidth utilization before you implement anything. This will help with determining if you have actually increased productivity.
Bottom Line, give the users 10 MB a piece, and the server 100 MB will allow 10 concurrent non-stop connections at a full 10MB before the server is at 100% cap.
NOTE: If you can support GB ethernet and you have the money, upgrade the server, then move the client's switch ports to 100 MB full, an the server at GB full, this will allow for the same scenario, but 10 times the bandwidth. I have to say that your network must be GB compatable before you move the server there. If you upgrade the card now, but the switches only support 100 MB,...........well, your server will be ready to go when you upgrade your switch.....
I can also tell you that if you use the 6509, experience with auto-negotiating Gig cards causes link flapping and intermittence. While upgrading the server's max throughput, you may hinder connectivity.
This will take some testing of the idea product...
Good Luck.
Second, can your cable plant handle Gig traffic? If not the card will (defaultly installed) auto-negotiate the next fastest connection, say 100 MB. I assume you already have that in current hardware. Below will discuss optimizing your netowrk and/or replacing components.
How many users do you have? Keep it simple, there may be no need to upgrade. These are the questions that you have to answer before you can even guess at your network's performance.
Are all of the clients on a hub or dedicated to a switch port? If they are dedicated to a switchport, then cut the client network cards to auto-negotiate and throttle the switch ports back half duplex. This will force each client workstation to 10MB. then put the server in the switch dedicated at 100 MB. This will keep any one machine from momopolizing the Server's resources. All users will see an improvement, unless you have 150 and one server.
If you can take the server and get it out of the broadcast domain with the clients, use helper-addresses on a router if it's also a dhcp server. This will reduce the amount of broadcast traffic the server has to process. (Not sure that an application server makes a good DHCP server, but hey you know your budget and your network size.
Take metrics of your current bandwidth utilization before you implement anything. This will help with determining if you have actually increased productivity.
Bottom Line, give the users 10 MB a piece, and the server 100 MB will allow 10 concurrent non-stop connections at a full 10MB before the server is at 100% cap.
NOTE: If you can support GB ethernet and you have the money, upgrade the server, then move the client's switch ports to 100 MB full, an the server at GB full, this will allow for the same scenario, but 10 times the bandwidth. I have to say that your network must be GB compatable before you move the server there. If you upgrade the card now, but the switches only support 100 MB,...........well, your server will be ready to go when you upgrade your switch.....
I can also tell you that if you use the 6509, experience with auto-negotiating Gig cards causes link flapping and intermittence. While upgrading the server's max throughput, you may hinder connectivity.
This will take some testing of the idea product...
Good Luck.
- Status
Similar threads
- Locked
- Question