The core architecture of this system is PCI-X 133 MHz/64-bit, which delivers 1 GB/s to the CPU. This bandwidth is split to several buses for the peripherals. What's relevant here is that there are two PCI-X buses:
Bus 1- 33 MHz/64-bit (266 MB/s) shared between VGA and the two Gigabit network controllers
This bus could run at at least 66 MHz, but falls back to 33 MHz because of the relatively low-end video card.
Each network controller can theoretically use up to 250 MB/s if running at maximum Gigabit speed in full duplex mode. In practice, we don't use that much, of course. I watched the network speed closely on Friday evening: it was about 30 MB/s outbound during MSS copy, and 12 MB/s inbound during DAQ - this comes nowhere close to the PCI bandwidth.
I do not know the actual throughput of the video adapter, but we only use 2D mode. A full screen update (3200x1200 pixels at 32 bits per pixel) would require a transfer of 15 MB. If you update 1/10 of this (one window) at 85 Hz, you are at 125 MB/s, and I think that would be really pushing it, because screen updates are only occasional, not streaming (unless someone decides to watch video...)
If network speed really becomes the bottleneck, we could install a video card with a faster bus interface.
Bus 2- 133 MHz/64-bit (1 GB/s) shared between the two SCSI adapters.
Each SCSI bus runs at 320 MB/s, and the sustained read rate from disk tops at somewhere around 75 MB/s. I do not this that the PCI-X bus of the SCSI controllers comes close to saturating.
Bottom line: I doubt we are running into the limits of the PCI-X buses at the current data rate.
A copy of this log entry has been emailed to: brdas, rom