7

Do the uplink ports on switches typically work okay as iSCSI ports? We're adding a 10gb iSCSI SAN, and want to get a combo switch (48x1gb & 4x10gb SFP+ uplink ports) and use the 10gb for the iSCSI SAN, while the 1gb are for a 1gb iSCSI SAN. We were told the uplinks don't provide buffering needed for iSCSI.

Is this specific to the switches used, and do some provide that needed buffering and some not on the uplink ports?

2 Answers 2

4

The definite answer to your question would be entirely implementation-dependent. "Uplink port" is nothing but a mere label - there is no consistent definition for what it is exactly. Typically it would be a high-bandwidth port, sometimes just a transceiver-less GBIC / SFP interface for additional media flexibility.

Whether it has less or more buffer space available than the other ports on the switch would be a question to ask the manufacturer. Whether the available buffer space would meet your specific needs, would be your very own task to determine in your lab setup with your set of load.

This being said, switch-side buffering is typically terribly overrated. Buffers are needed to compensate for different port speeds so the host on the faster link would be able to send data to the slower links in bursts which the switch's buffers would equalize. But since the storage controller has lots of buffer space available to himself with flow control mechanisms at different layers of the transmission (Ethernet if you enable it, TCP in any case), the switch buffers technically just would need to be large enough to hold the amount of ([uplink port link speed] * [number of uplink ports] / [slowest link speed]) + 1 frames in order to be able to keep the slower downlinks saturated at all times. With 2 10G uplink ports connected, 1G slowest links and jumbo frames enabled this would amount to just under 200 KB of buffer space. Due to over-engineering and implementation constraints (e.g. a fixed relation of buffer space to input-output port pairs) you would see significantly larger total buffer space declarations in switches' data sheets.

1

I fully agree with syneticon-dj's answer.

I think whoever made that statement about buffering is parroting something dating back to the 10/100 MB switch days. In those days there were a lot of shitty designed switches around, even from reputable vendors that didn't have proper buffering in the first place.

When 1 Gb interfaces became normal having 512K to 1 MB buffers per interface became more or less a standard feature. 10 G interface typically have 1 to 4 MB buffers each. (Or a cluster of 2 or 4 interfaces shares a common 4, 8 or 16 MB buffer.)
I have never encountered unbuffered 10 G interfaces.

Any modern switch capable of handling 10 G interfaces should have ample internal buffering capacity to have these 10 G interfaces used for iSCSI.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .