Q: dual nic configuration
- Thread starter msents
- Start date
Adaptive Load Balancing is what intel calls it. For an example of how to do this on win2k, using paired intel pro 100+ NICs, look at this Dell support doc.
http://support.dell.com/us/en/kb/document.asp?DN=1045120
Any suggestions on Fast EtherChannel (to cisco hardware) for OSX users? There must be a way to get faster connectivity out of these boxes.
A quick one: Can anyone explain why AFP hogs more than 50% of the cpu (c.f. >top) with single workstation connected at 100 megabit on same speed switch (client is emptying trashed items from share)? Server is B&W G3 w/^256RAM. Comparable Dell box (faster processor of course) seems to handle it better. So far I am disappointed with performance of this HW/SW/Price combo.
Lastly: Can anyone tell me a reliable inexpensive HW or SW raid solution for X?
http://support.dell.com/us/en/kb/document.asp?DN=1045120
Any suggestions on Fast EtherChannel (to cisco hardware) for OSX users? There must be a way to get faster connectivity out of these boxes.
A quick one: Can anyone explain why AFP hogs more than 50% of the cpu (c.f. >top) with single workstation connected at 100 megabit on same speed switch (client is emptying trashed items from share)? Server is B&W G3 w/^256RAM. Comparable Dell box (faster processor of course) seems to handle it better. So far I am disappointed with performance of this HW/SW/Price combo.
Lastly: Can anyone tell me a reliable inexpensive HW or SW raid solution for X?
Attachments
the Disk Utility makes a good RAID.
I wouldnt put mission critical work on it though. Traditional
RAID requires a number of SAME sized disks. I have put 6 SCSI disks, none of them match in size, and made a Striped Raid from it.
This shouldnt have been possible. It appears you can make a striped raid from 2 disks as well.
This also shouldnt be possible.
Any other experiences with this?
I wouldnt put mission critical work on it though. Traditional
RAID requires a number of SAME sized disks. I have put 6 SCSI disks, none of them match in size, and made a Striped Raid from it.
This shouldnt have been possible. It appears you can make a striped raid from 2 disks as well.
This also shouldnt be possible.
Any other experiences with this?
AFP on ten server is not as fast as I would have expected.
I have found some minor ways to increase speed (would you believe that AFP runs 7% faster in my testing when I turned off the screen saver?) but overall I am disappointed with the results. I don't know yet if the problem is with the hardware or the software, if anyone has any real results, please let me know. I do know that I have several win2k boxes that run as fileservers at a minimum 4X faster than the OS X server box I just built (using the Helios LanTest utility to benchmark the systems). There are significant differences in hardware though. Most of the bottleneck so far is the network, beyond that it looks like processor and hard drive but I think the problem is software utililization of resources rather than hardware itself. Doing a 30Mb contiguous file write test I get a sustained average transfer rate of 6766.92 Kb per second on this OS X box which is fine for 1 user hitting it, but I have significantly higher transfer speeds on Win2K with 75 simultaneous connected users.
I do have one idea that I may implement out of curiousity: I want to replace the built-in ethernet card with one that has a better reputation, for example a Farallon card. The built-in and Apple branded cards have historically given me trouble, whether from hardware or drivers.
Looks like we aren't getting anywhere on the Adapter Teaming question, perhaps someone has some test results for the different brands of NIC under OS X? I'm curious especially about Farallon and Asanté.
Does anyone know if installing Darwin gives you AFP services?
I have found some minor ways to increase speed (would you believe that AFP runs 7% faster in my testing when I turned off the screen saver?) but overall I am disappointed with the results. I don't know yet if the problem is with the hardware or the software, if anyone has any real results, please let me know. I do know that I have several win2k boxes that run as fileservers at a minimum 4X faster than the OS X server box I just built (using the Helios LanTest utility to benchmark the systems). There are significant differences in hardware though. Most of the bottleneck so far is the network, beyond that it looks like processor and hard drive but I think the problem is software utililization of resources rather than hardware itself. Doing a 30Mb contiguous file write test I get a sustained average transfer rate of 6766.92 Kb per second on this OS X box which is fine for 1 user hitting it, but I have significantly higher transfer speeds on Win2K with 75 simultaneous connected users.
I do have one idea that I may implement out of curiousity: I want to replace the built-in ethernet card with one that has a better reputation, for example a Farallon card. The built-in and Apple branded cards have historically given me trouble, whether from hardware or drivers.
Looks like we aren't getting anywhere on the Adapter Teaming question, perhaps someone has some test results for the different brands of NIC under OS X? I'm curious especially about Farallon and Asanté.
Does anyone know if installing Darwin gives you AFP services?
Well, I tried a couple of configurations with additional NICs and the results are interesting. Installed an Asanté and a Farallon card. Set each card to dhcp (have a dhcp server up) and forced each one to a specific ATalk address in a specific range (no seed router). None of the cards actually took the atalk address. AFP service is running on one IP address but is configurable via any of the ones assigned. All machines on network connect to same address though. Having some difficulty connecting to the machine (looks like it fails via IP, falling back to appletalk type thing but then it connects via ip so perhaps there is some kind of contention for the connnection? The Asanté card is dark after a reboot and never lights again. Replaced all of the NICs with 3 Farallon cards. Plugged in all 3 to the switch. No built in, no Asanté on this test. Each gets an IP address, each is manually assigned an Atalk Address. Everything works, no more lag while attempting a connection, only one IP address actually gets the share from client perspective. I think I am wasting time here. Turned off the other interfaces and ran LanTest, getting faster throughput but not sure why. I'd like to blame the built-in adapter but it is probably just config changes. What have I learned? It looks like you can install additional adapters but all of the services must run from the same one when configured via the GUI. One more interesting quirk. I disabled two of three NICs via the System Preferences tool. One went dark shortly after, one stayed lit. Verified that I had actually disabled both. Yep. One dark, one lit. Traffic too. Wonder what it's talking about? Maybe I'll sniff around on this next.