Re: multi-port NIC cards
--- martin [EMAIL PROTECTED] wrote: Hi. I just ordered both the Mikrotik Routerboard 44 ($89) and the Soekris lan1641 ($95). Both 4-port NIC boards. I'll let you know how the perform. I'm also puzzled by the claims of performance issues and saturating the bus PCI bus previously mentioned as the original PCI (33MHz) has approx. 1 Gbit performance and these cards have 4x100 Mbit chips and therfore will only use 400 Mbits maximum of the 1 Gbit bus. Is someone confusing bits and bytes ? Regards...Martin
Re: multi-port NIC cards
Daniel Ouellet skrev: May be good, but the bus is PCI only if I am not mistaken looking at the spec. Not even PCI Express or PCI X, so it would be interesting to see, but if you are concern about congestions with the Intel one, may be this would be saturating the bus at 33MHz, or may be it might go at 66, but sure not 100 or 133 however. I saw some others, but none that support PCI Express as a minimum however. So, I discarded them. i haven't tested any 4 port nic's whatsoever yet, and don't know much about these things, but isn't the theoretical throughput of the 33 MHz 32-bit pci bus around ~1 Gbit/s? so, assuming the system is dedicated to routing, why would a theoretical maximum of ~0.4 Gbit/s be so hard to handle, especially as most of it should stay on the internal pci bus of the nic? kindly kami petersen
Re: multi-port NIC cards
33 Mhz * 32 bits = 1 056 000 000 bits per tick, 1 056 000 000 / 10^6 (1 megahertz = 10^6 ticks per second) = 1 056 megabits per second 1 056 / 8 = 132 megabytes per second It should actually be 100/3 Mhz. kami petersen wrote: Daniel Ouellet skrev: May be good, but the bus is PCI only if I am not mistaken looking at the spec. Not even PCI Express or PCI X, so it would be interesting to see, but if you are concern about congestions with the Intel one, may be this would be saturating the bus at 33MHz, or may be it might go at 66, but sure not 100 or 133 however. I saw some others, but none that support PCI Express as a minimum however. So, I discarded them. i haven't tested any 4 port nic's whatsoever yet, and don't know much about these things, but isn't the theoretical throughput of the 33 MHz 32-bit pci bus around ~1 Gbit/s? so, assuming the system is dedicated to routing, why would a theoretical maximum of ~0.4 Gbit/s be so hard to handle, especially as most of it should stay on the internal pci bus of the nic? kindly kami petersen
multi-port NIC cards
Hello. Can anyone recommend a good multi-port NIC card e.g. 4-port, that works OK on OpenBSD with a good source supplier. Regards...Martin Just $16.99/mo. or less. dsl.yahoo.com
Re: multi-port NIC cards
Can anyone recommend a good multi-port NIC card e.g. 4-port, that works OK on OpenBSD with a good source supplier. This question was debated a few times in the archive already. So, far there isn't one great card that works very well that still available to purchase new these days. SK based were best, but not available anymore. Intel have the Pro 1000MT, but you need to run the bsd.mp to not get overwhelm by interrupts even on a single processor server. That card works, but not as well as it should really! I would just OK with bsd.mp, but not under very heavy load, but will do what you want for lower demanding setup as long as you DO run the bsd.mp kernel. So, far I haven't found one that is still available to purchase new these days and will provide the same efficiency as older cards were able to do! (: Very sad but true! I sure hope this change soon, but that's where we are now, at a minimum, that's where I am anyway. Daniel
Re: multi-port NIC cards
On a related subject and please forgive any ignorance on my part, how would the interrupt load compare, between a multi-port NIC and the same number of ports via individual single port NICs? For example, a firewall with one WAN port and three LAN ports. One LAN (and of course the WAN port) port would see 'heavier' use for a family's worth of surfing, another would be a DMZ (WAN - DMZ only) and the last would be a highly restricted and far less used 'other' route into the DMZ, but only from the LAN side and NOT the WAN side. Traffic would ultimately be bottle-necked on the WAN side, with an 8Mbit/384Kbit DSL connection with a little AltQ if and when it is required. Regards, Craig On Sun, 2006-01-01 at 16:21 -0500, Daniel Ouellet wrote: Can anyone recommend a good multi-port NIC card e.g. 4-port, that works OK on OpenBSD with a good source supplier. This question was debated a few times in the archive already. So, far there isn't one great card that works very well that still available to purchase new these days. SK based were best, but not available anymore. Intel have the Pro 1000MT, but you need to run the bsd.mp to not get overwhelm by interrupts even on a single processor server. That card works, but not as well as it should really! I would just OK with bsd.mp, but not under very heavy load, but will do what you want for lower demanding setup as long as you DO run the bsd.mp kernel. So, far I haven't found one that is still available to purchase new these days and will provide the same efficiency as older cards were able to do! (: Very sad but true! I sure hope this change soon, but that's where we are now, at a minimum, that's where I am anyway. Daniel
Re: multi-port NIC cards
Craig McCormick wrote: On a related subject and please forgive any ignorance on my part, how would the interrupt load compare, between a multi-port NIC and the same number of ports via individual single port NICs? You don't really have something to compare with. The process is way different how the interrupts are handle in the bsd and bsd.mp kernel. It's driver related as far as I understand it. If Intel have better available documentations may be it would be handle better, but it is not the case here. That's what I understand anyway and I am sure if I am wrong, someone will put me back in strait line here. If you can install multiple SK based cards for example and you are looking to get very efficient handling, you would be better served with that. But looking at your requirements, I may be wrong, but I would guess that even the Intel one would be fine for what you want to do. Doesn't look like each port would handle 50Mb/sec minimum or something like that. Your choice, but if you really want best of brand, stick with SK, if you need integrations, then in your case I would think Intel would be fine as well, just run bsd.mp to help. Hope this help you anyway. Daniel
Re: multi-port NIC cards
--- Daniel Ouellet [EMAIL PROTECTED] wrote: Can anyone recommend a good multi-port NIC card e.g. 4-port, that works OK on OpenBSD with a good source supplier. This question was debated a few times in the archive already. So, far there isn't one great card that works very well that still available to purchase new these days. SK based were best, but not available anymore. Intel have the Pro 1000MT, but you need to run the bsd.mp to not get overwhelm by interrupts even on a single processor server. That card works, but not as well as it should really! I would just OK with bsd.mp, but not under very heavy load, but will do what you want for lower demanding setup as long as you DO run the bsd.mp kernel. So, far I haven't found one that is still available to purchase new these days and will provide the same efficiency as older cards were able to do! (: Very sad but true! I sure hope this change soon, but that's where we are now, at a minimum, that's where I am anyway. Daniel Just found this. http://www.routerboard.com/rb44.html Might just buy one and try it out. Regards...Martin
Re: multi-port NIC cards
martin wrote: Just found this. http://www.routerboard.com/rb44.html Might just buy one and try it out. May be good, but the bus is PCI only if I am not mistaken looking at the spec. Not even PCI Express or PCI X, so it would be interesting to see, but if you are concern about congestions with the Intel one, may be this would be saturating the bus at 33MHz, or may be it might go at 66, but sure not 100 or 133 however. I saw some others, but none that support PCI Express as a minimum however. So, I discarded them. That doesn't mean this is not a good one. (: I didn't test it, so I can't talk knowing the results.
