[rescue] Bus Speed v I/O rules of thumb?
Francisco Javier Mesa-Martinez
lefa at cats.ucsc.edu
Fri May 17 14:23:54 CDT 2002
On Fri, 17 May 2002, Leslie Connally wrote:
> With all this talk of overloading Joshua's little SS10 w/ sbus cards, can
> anyone give any sort of rules-of-thumb for I/O and busspeed ? esp for those
> us us who are not engineers. What other issues are involved? Is there not
> CPU load from the I/O device itself..whether ethernet or SCSI? How the hell
> does MHz relate to MB/s anyway (bet I dont want to ask that .. do I?)
> Remember.. keep it simple..
Mhz relates to MB/s in the following fashion:
1Mhz = 1x10^6 1/s, i.e. 1 million cycles per second (hz = 1/s)
So assuming that our bus is running at x Mhz, and it has a bus width of y
bytes. We can calculate the speed of the bus simply by:
Busspeed = (x Mhz)*(y Bytes) = (x*y) (M)(Bytes)(1/s) = x*y MB/s
This is given a single cycle the bus can transfer y bytes, and the bus
performs x million of cycles in a secod.
Note that this is the peak number, which usually does not reflect the
actual transfer rate for the bus. Since there are certain issues like
protocol overhead (i.e. the bus is a shared medium so a protocol is needed
for devices to comunicate using the bus and limiting contention issues).
An that will take a % out of your bandwidht. I.e. if I need to tranfer a
chunk of Z bytes, I need to spend sometime trying to get the bus for me...
send the Z bytes (or part of them in case the bus only allows a certain
packet size, so the Z bytes will need several transfers to actually be
transmitted) and then release the bus. So usually the overhead has a
significan effect in how much of the theoretical bus bandwidth you get to
use.
Of course a bus is a shared medium, so the more stuff you hang to it, the
more the performance is going to suffer. That is why putting 2 scsi boards
+ 2 FDDI + a fast ethernet + .... on a single bus is not a good idea
Hope this helps.
More information about the rescue
mailing list