Digging through the backlog, and found this unanswered... Joe S wrote: > Since some quad nics share 1 interrupt, what kind of performance impact > would I be dealing with versus using 4 indiviual nics? > > Debating wehter to use a Phobox P430TX quad dc nic or individual fxp0 nics.
Just test it and find out. On the criteria you specify, shared vs. individual IRQs, you are unlikely to see any difference. (it was believed that shared IRQs are faster, but the difference was very slight last time someone actually measured, I do believe.) It depends on what machine you stick that Phobos card in, anyway. The Compaqs I've stuck 'em in give 'em shared IRQs, (Celeron 300) > ppb1 at pci0 dev 13 function 0 "DEC 21152 PCI-PCI" rev 0x03 > pci2 at ppb1 bus 2 > dc0 at pci2 dev 4 function 0 "DEC 21142/3" rev 0x41: irq 11, address > 00:60:f5:08:54:2c > lxtphy0 at dc0 phy 1: LXT971 10/100 PHY, rev. 1 > dc1 at pci2 dev 5 function 0 "DEC 21142/3" rev 0x41: irq 11, address > 00:60:f5:08:54:2d > lxtphy1 at dc1 phy 1: LXT971 10/100 PHY, rev. 1 > dc2 at pci2 dev 6 function 0 "DEC 21142/3" rev 0x41: irq 11, address > 00:60:f5:08:54:2e > lxtphy2 at dc2 phy 1: LXT971 10/100 PHY, rev. 1 > dc3 at pci2 dev 7 function 0 "DEC 21142/3" rev 0x41: irq 11, address > 00:60:f5:08:54:2f > lxtphy3 at dc3 phy 1: LXT971 10/100 PHY, rev. 1 (I'm sure not all Compaqs do that, but I seem to have a pile of P90 through PII-450s that seem to only allocate one IRQ to the PCI bus, only thing in common between them is the Compaq badge on the front. This machine had at least two other devices sharing that same IRQ 11. Put four fxp cards in there, you will see them all on IRQ 11, too). however, the same kind of card in this Dell does not: (Celeron 400) > ppb1 at pci0 dev 13 function 0 "DEC 21152 PCI-PCI" rev 0x03 > pci2 at ppb1 bus 2 > dc0 at pci2 dev 4 function 0 "DEC 21142/3" rev 0x41: irq 9, address > 00:60:f5:08:54:20 > lxtphy0 at dc0 phy 1: LXT971 10/100 PHY, rev. 1 > dc1 at pci2 dev 5 function 0 "DEC 21142/3" rev 0x41: irq 10, address > 00:60:f5:08:54:21 > lxtphy1 at dc1 phy 1: LXT971 10/100 PHY, rev. 1 > dc2 at pci2 dev 6 function 0 "DEC 21142/3" rev 0x41: irq 5, address > 00:60:f5:08:54:22 > lxtphy2 at dc2 phy 1: LXT971 10/100 PHY, rev. 1 > dc3 at pci2 dev 7 function 0 "DEC 21142/3" rev 0x41: irq 11, address > 00:60:f5:08:54:23 > lxtphy3 at dc3 phy 1: LXT971 10/100 PHY, rev. 1 (I'll spare you the dmesg from the five 4port NIC dell I set up. Needless to say, IRQs were shared. :) So, if you have to switch between different HW to get shared vs. "personal" IRQs, you have changed a lot more variables... You will see a much bigger difference about the different chips you have (dc vs. fxp, or even which fxp version you get). Probably even bigger than that would be the design of the machine -- if you have a machine where each individual NIC is on a separate PCI bus, you may see better performance than having them all on one bus (like you would on the quad-port card). HOWEVER... *IF* you really have a situation where you would really see a performance difference between the Phobos and four fxps, you probably need to say, "WRONG CARD" and grab some good gigabit NICs instead. This isn't auto racing, no one will ever notice a percent or two difference in performance. If you got to use a stopwatch or benchmark to see the difference, it's wasted effort. To get any real difference, you will need some (even) better NICs. Nick.
