1. Home
  2. Ethernet LAN
  3. Maybe a simple question about switching

Maybe a simple question about switching

Hello everyone, Our office servers room, among many other machines, have some oracle databases that, for some unknown reason to me (maybe security issues) were configured in a LAN segment different from the rest of the others servers - the oracle servers are in a 172.19/16 subnet, and all the others are in a 172.20/16 subnet. To handle the traffic between them, the last admin set up a linux firewall with two 100Mb NICs , and enabled packet forwarding into the OS. I´m not that good on network concepts, but as far as I understand this machine is acting like a router between two network segments (is it right)? The problem is, the databases are being more and more used by the other servers (java applications, web servers, etc), and I´m seeing that the

100Mb connection between the NICs is becoming a big bottleneck. One simple solution would be buying gigabit NICs, but a friend told me it would bring almost no gain, since the firewall machine is a common 32bit Intel server, with 33Mhz PCI connectors (no 66Mhz/100Mhz PCI-X to handle all the extra bandwidth). So, I tought about some options that I would like to know which one of them would work best , hoping that I´m not saying something stupid or wrong: 1) Buying a new server with PCI-X slots to fully use the extra bandwidth of the gigabit NICs (not cheap); 2) Simply use a gigabit switch configuring it with a class A mask (172/24), so it would "listen" to both subnets and share packets betweem them; 3) Use a router (bridge) with two gigabit ports, one on the 172.20/16 and the other on the 172.19/16 subnets (does this thing even exist?)

I hope someone can help me to understand this, and point me to the right direction.

Thanks a lot,

Luis Derani.

Reply to
Derani
Loading thread data ...

Most "2001" machines can transfer 300-400 Mb/sec RAM-to-RAM via gigabit. Going to gigabit on them won't give you a billion bits per second between them but it can give better performance than 100TX and it's a lot cheaper and less complex than link aggregation schemes.

Reply to
J. Clarke

Derani wrote in part:

[snip]

Why do you think it's the network bandwidth and not some other component? 100baseTX can pass 12+ MBytes/s . You need pretty fast disks and servers to keep up with that.

In any case, timed `ifconfig`s on the Linux router will show the bandwidth being used.

As much as I like Linux, it needs careful optimization when used for a high-bandwidth router. You can test your install by running `ttcp` on both sides.

Since switching has ended "shared bandwidth", there's no point in going up to gigabit unless you replace end machines. Gigabit can deliver 120+ MB/s, which is faster than RAM speed on 2001 machines.

-- Robert

Reply to
Robert Redelmeier

Agreed, and an increase to even PCI gigabit can give a useful boost. But only if 100 is actually saturated. I strongly suspect it is not, most likely due to Linux config issues.

-- Robert

Reply to
Robert Redelmeier

Hello Robert and J.Clarke,

thanks for the answers. The fact I tought it could be a bandwidth issue is because I take care of all the setup now (the oracle db, the java app on the other side, and the linux firewall machine and links), and lately the users started to use the system more heavily, and at the same time the system started to slow down as a whole. I´m trying to speed it up on all the possible ways (optimizing oracle, and helping the developers write better queries) and made some good advances lately, but a topic that I don´t have much experience is the network part - the only place that I still didn´t change anything. So, I´m gonna do some reading to better understand how to optimize the network parameters of the firewall and tcp/ip stack of the linux box as it is now (with PCI 100Mb NICs) and also try to change to gigabit NICs to see if it makes a difference even with the standard PCI bus.

[]´s!
Reply to
Derani

Good. See also if that firewall is slamming disk. mount NOATIME might help.

Running `ifconfig`, waiting 10 seconds and running it again during peak loads will give you by subtraction exactly how much bandwidth is going across the router. That will give you a good idea whether hardware will help.

-- Robert

Reply to
Robert Redelmeier

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.