Dual NICs for redundancy, howto?

[blakerwry]

I was wondering if anybody had suggestions on the best way to implement dual NICs for redundancy purposes.

I am planning on putting together a Linux server for my counter strike clan. Since I have a lot of 10baseT NICs I was going to put 2 of em in this server for redundancy.

I have my choice of dual 3com 509 NICs or HP Ethertwist NICs... or I also have one each of Linksys LNE 2000 and SMC EZcard 10(model 1660T) NICs.

These NICs are all ISA 10Mbit cards.


So, if i was wondering if it's a good idea to use 2 NICs... if so, should I use the same model or should I use different models... Is there any special configuration I'll need to use or will it basically work automatically? Will the machine have 2 IP's or just 1?

The IP of a halflife server is actually pretty important because the way HL client was implemented, people connect to the IP instead of a URL... a static IP is pretty much a necessity.. probably a single static IP...(im not sure HL server supports being run over 2 seperate IPs)

If the HL server doesn't support being run on 2 IP's simultaneously, is it still a good idea to include a second NIC? This way the second can be configured to take over for the 1st in the event of failure without 1) physical access to the server and 2) without restarting the server...

what do you guys think?

 

[Stereodude]

You can't in Windows unless the drivers supports it. You need two identical NICs and drivers that support redundancy.

Stereodude

 

[Bozo]

In W2k you can install 2 NICs and the disable one in software. If one fails, you're just a couple of clicks away from restoring.

3Com does make NICs that have redundancy built in but they are pricey.

Bozo

 

[James]

I don't know of any ISA NICs that support failover natively through their drivers under Unix. You're better off going through the OS.

You're using Linux so yu have a number of options - the code from Oracle at http://otn.oracle.com/tech/linux/open_source.html is probably a good start, or just Google for "NIC failover Linux" or something similar (that turned up a bunch of solutions, including a brief overview at http://lists.suse.com/archive/suse-oracle/2000-Aug/0095.html ).

 

[blakerwry]

I found out that solaris offers something called alternate pathing, but this is not an automtic fix... ie, the admin must enter in a command to have 1 NIC take over for the other...

so i don't see the point of using it when I could just leave 1 NIC's IP address field blank and fill it in with the correct IP in the event that the 1st on fails

 

[Mercutio]

Spoken like someone who has never been awakened by the dreaded 3:00AM pager.

 

[P5-133XL]

I'll ask the really dumb question - Why can't you make each the others default gateway. thereby, if the message fails to get through it will be sent via the other network card. If the network cards are attached to a router or switch then they will take care of the details of getting the message to the proper card at the other end. Specificly the router will route via the destingation IP address and the switch will do the same for the destination MAC address.

Maybe, I'm blowing smoke but by my understanding, it should work as a fail-over.

 

[blakerwry]

Spoken like someone who has never been awakened by the dreaded 3:00AM pager.

true... but this will not be for work... actually.. i will probably be playing on the server at 3:00am... ;-)

I hope to have remote webmin access or something on the server so it shouldn't be too bad... should be able to fix things from home.

In a worse case senerio it can wait till morning when someone gets to the server (they'll be there normal business hours every weekday.. maybe weekends).


hmm.. that makes me wonder, what's the best way to restart a frozen server remotely? is there a way?

 

[P5-133XL]

They make cards to remote start servers

 

[blakerwry]

If I made each card's gateway point to the other card then wouldn't they both go down if only 1 card failed?

Even if it didn't, i don't see how 1) this would prevent automatic failover and 2) would be any better then leaving one of the NICs basically unused for the server and setting it up in case of a failure of the primary NIC.

 

[P5-133XL]

The GW address is used when the server can't get through. Thus setting the GW to the other card would mean the server would send the down cards packets to the other card. That card would then send the data down the line to the switch/router. The failure mode is easily testable.

With the GW method, both NIC's can be used doubling your potential throughput

 

[P5-133XL]

The GW address is used when the server can't get through. Thus setting the GW to the other card would mean the server would send the down cards packets to the other card. That card would then send the data down the line to the switch/router. The failure mode is easily testable.

With the GW method, both NIC's can be used doubling your potential throughput.

 

[P5-133XL]

The GW address is used when the server can't get through. Thus setting the GW to the other card would mean the server would send the down cards packets to the other card. That card would then send the data down the line to the switch/router. The failure mode is easily testable.

With the GW method, both NIC's can be used doubling your potential throughput.

 

[honold]

the gateway trick would not work at all.

of more immediate importance, do you have raid1, dual power supplies, and a ups on this system? those components are infinitely more likely to fail unless you've got your pci bus cranked.

just get a single intel eepro/100s and be happy/fast/stable. once you implement the above stuff, you are allowed to worry about HA on your nics

 

[blakerwry]

heh, ok.

I was basically planning on giving a couple hundred $$ of old equipment (working 440BX w/ pIII system) so my clan could put it in a rackmount server, unfortunately I just found out we are limited to a 1u or 2u rack... so my slot pIII ain't going to work.

We're trying to find some good deals on a case and internals.. we've been looking at refurb/used dells and also looking at buying a case and building the system from new/donated parts.

We're looking at p4 or athlon systems or possibly dual pIIIs.. we have a p4 and several athlons that could be donated if need be... what do you guys suggest I look into? any good places to find a cheap rackmount server in 1/2u form?

 

[honold]

http://www.servercase.com

 

[blakerwry]

looks like anywhere we go we'll be spending ~$200 on a case/PSU....

what do you think about somethign like this?

Seems like an OK price for case, floppy, slimline CD, PSU and an iNTEL 845 p4 mobo.

we have a 1.7ghz p4 we could pop in there... and if we later needed it we could goto a 400mHz FSB 2.6gHz northwood.


If we buy the case and parts seperate, what kind of PSU should we get? are there any name brands to look for (or avoid) and waht type of wattage rating are we going to need for a p4/athlon system?

 

[honold]

looks okay to me, supermicro certainly has a good rep as well

enermax, sparkle, and seventeam are all big names in rackmount psus. 400+w should be more than enough.

 

[blakerwry]

400 watt?

It seems like most 1u cases come with 250-300 watt units... I was looking for which I want? or is neither good?

 

[honold]

what kind of disks are you using, and how many?

 

[timwhit]

You definately don't need a 400W power supply. You would be fine with 250-300W.

 

[blakerwry]

Just a single IDE disk 7200rpm. ... 1 full size CD-ROM drive... we'll probably get a single CPU athlon/p4 mobo and 1GB of RAM.

 

[honold]

300 should be plenty then

 

[time]

200W would be plenty - for a single midrange CPU and one drive.

 

[honold]

expansion should be accounted for - rack psus are more expensive and not exactly the kind of thing one has laying around

 

[time]

So true.

 

[Dïscfärm]

...what do you think about somethign like this?

If you are using it to host something as un-serious as gaming, I'd say it is overkill -- even though the price is certainly reasonable.

These Supermicro boxes are designed for dead-serious rack'em stack'em 24x7 business and industrial server applications.

However, if you believe that gaming is serious business (I don't, but then again I NEVER play computer games), then why not!

http://www.supermicro.com/Product_page/product-s.htm

 

[blakerwry]

One of the senior clan members works at a place with an OC3 connection, he is allowed to put our game serer on it if he works a couple extra hours every week. 1 of the conditions is the server *needs* to be in a 1U or 2U case.

The cheapest rent-a-server style solutions we could find was $600 for 6 months (I believe this was for three 16 person game(Half Life) servers. And HL is an easy task.. a game like Battlefield 1942 costs about $8-12$ per player monthly there's a 16 player minimum... 24 if you want to compete.

We've already raised between $200-300 dollars in donations and we already have a donated CPU, HDD, and CD-ROM.

Considering this, we could build a rackmount that will 1) cost the same or less as a hosting solution 2)will not cause us to have to pay reaccuring fees 3) will support more servers and more games and 4) we will have something we own that we can use for whatever we want.

So basically this will save us alot of money, even if we do get the overkill supermicro setup. I'd rather go overkill on the essentials and upgrade items like the CPU/RAM later as our needs change.

 

[Dïscfärm]

Well, seeing your requirements and your (his) hosting predicament, this is NOT really overkill.

 

[blakerwry]

Do you think we should go with the name brand supermicro setup for $500 or should we buy a case ourselves for $200 and a regular micro ATX motherboard for another $100?

I was actually considering this: A7N266-VM As e_dawg gave it his approval as the best bargain board and it looked like it was very good from the newegg customer comments.
$72.00...


The DDR would probably be cheaper than the PC133(used by the supermicro setup) right now... we have both a 1.7gHz p4 or a 1700/1800+ AthlonXP that people are willing to donate.


I was at first leaning towards the supermicro just for ease of setup and a less hassle factor... probably also get better support... However, after pricing the case, PSU, and mobo the $200 we could save made me lean towards the seperate case and regular PC mobo option...

I have to say I'm torn... but if a regular PC mobo would work for us then I'm all for it.


I should say that I've hosted Half life game servers off my Soyo Dragon+, ak32L, and SB51G... they all performed fine and were very stable...

Current uptime on my SB51G is 20 days 11 hours 1 minute... I've done some upgrades/modifications to the 3 servers(1CS, 1 TFC, and 1 DoD) I'm hosting at my house so the servers have gotten restarted several times... the only stability problem I've had is that the CS server stopped responding to connections once... simply restarted the service and it was back up in about 10 seconds.


So there's nothing that leads me to believe that it can't be done on regular PC hardware.. but I'm just wondering about long term use of a PC motherboard in a rackmount 24/7 situation.

 

[honold]

a normal board will likely work just fine. i would just make sure that you use cool drives and cpus - these will be stacked and confined.