Shared bandwidth

[Canon Bob]

It's not uncommon for MoBo architecture to share out resources between different devices. When PCIe slots share bandwidth with SATA, USB or M2 connectors, is that bandwidth reduced by simply having a module or device installed/connected or does the sharing only affect performance if both items are being accessed simultaneously?

Cheers....Bob

 

[dprees]

It depends. Each motherboard will have a fixed number of PCIe lanes, which are allocated to various sockets. In an ideal world, all the sockets would have their own dedicated lanes. However, there are rarely enough of these lanes, and so the board manufacturers have to make choices. One approach is simply to put sockets onto the board which just match the number of lanes available. An alternate approach, to provide flexibility for the differing needs of the purchasers, is to have a subset of the lanes be connected to more than one socket. Plugging a device into one of the sockets "takes over" those shared lanes, and the other socket becomes non-functional (whilst the other socket is occupied).

For certain subsystems on a motherboard, the bandwidth provided by lanes connected to them will be shared between multiple connected devices -- this happens with USB, for example. For this scenario, it matters that normally any one device is unlikely to make use of the bandwidth all of the time (infrequent transient peaks of usage), and therefore sharing it out makes sense.

 

[Canon Bob]

Thanks David. I understand that sometimes A is disabled by B....no problem there. But if A shares b-w with B then does A only ever get half the B-W even if B is not being used (B is perhaps an archive drive for example) or does the chipset only reduce A's B-W if B is called into action?

 

[dprees]

My understanding is that when bandwidth is shared (perhaps several SATA drives all connected via the same set of lanes) then all of the bandwidth is available to any device which is operating, if all the rest are not doing anything.

 

[Canon Bob]

My understanding is that when bandwidth is shared (perhaps several SATA drives all connected via the same set of lanes) then all of the bandwidth is available to any device which is operating, if all the rest are not doing anything.

Thanks again David....that's hat I was looking for.

Initial logic would point one to using shared bandwidth with lesser used items but it's perhaps better to pair up items that are unlikely to be used at the same time so that they benefit from maximum performance when they are used.

 

[dprees]

Bob,

The shared bandwidth approach makes sense also when any use of the bandwidth is likely to be very short in duration, followed by a pause before being used again. This allows multiple devices to interleave their traffic over the lanes, each effectively enjoying the benefit of maximum bandwidth, with only rare exceptions when a collision occurs. This is a perfect match for backing storage devices -- they are polled to read/write a block of data, and then they pause, due to a delay for some reason (data being processed from RAM, moving read/write heads on disk, etc.). The same would be true for USB devices, too.