[ Double-click this line for list subscription options ]
Andrew E. Mileski wrote:
> eMail2Me wrote:
>>
>>>BTW--my experience has been that most 33Mhz PCI busses can't keep the
>>>1Gbps ethernet NICS full, and that they max out around 350-500 Mbps
>>>(including overhead). Has anyone else noticed this?
>>
>> Currently, I don't have a 1000Mbps LAN connection. However, since the PCI
>> is a 32-bit bus (address/data) and clocked @ 33Mhz, therefore it should
>> be able to deal with a 32 x 33Mbps data on its path, theoretically
>> speaking. About three months ago, I ran into someone who told me that his
>> computer's SCSI hard drive got chocked (overwhelmed) with fast and vast
>> amount of downloaded data piping through a giga switch connection.
>
> Being pedantic, a PCI read is 3 cycles and a write is 2 cycles, so
> 33 MHz / 3 cycles * 32 bits = 342 Mbits/sec on read
> 33 MHz / 2 cycles * 32 bits = 528 Mbits/sec on write
> Of course a burst is even quicker, though they are usually 64 cycles
> or less long. Assuming optimal timing:
> 33 MHz / 66 cycles * (64 * 32) bits = 1024 Mbits/sec on read
> 33 MHz / 65 cycles * (64 * 32) bits = 1039 Mbits/sec on write
> Anyways one NIC will completely monopolize the PCI bus.
>
> Most OSes transfer to main memory then out again. This halfs the
> effective rate. Luckily hard drives are slower.
------- End of forwarded message -------
---
* Origin: [adminz] tech, security, support (192.168.0.2)
generated by msg2page 0.06 on Jul 21, 2006 at 19:04:37
search: