Best pratcice 100/1000

[OliveSico]

Hi all,

Looking on some of my Procurve switch, i recently found that some users' computers are connected at 1Gb link (current mode). All switches (48 ports) are trunked with 2 1Gb copper or fiber link to their gateway.

I wonder if i should force all ports of the switces at 100mb Fdx or let the computers auto-negociate the link.

Thanks for any reply :)

[parnassus]

Why you should do that?

I mean: Are you worried that your access devices (PCs) correctly connected to their respective Switch at 1Gbps (if auto-negotiation is "On" and 1Gbps is the result of that auto-negotiation you are good) are saturating the uplinks to your gateway? well...are you sure about that or is just a worry? you have to think about the traffic concurrency...often having forty or more ports connected at 1Gbps doesn't necessarily mean that those ports are managing (and so are full used) 1Gbps constantly.

The same couldn't be said about the uplink of 2Gbps (2x1Gbps in LAG?) since the uplink carries the whole switch traffic that need to go to the upstream device (your gateway, in this case)...so much depends on your specific network topology, local Server host positioning and network usage (internal and external traffic) by your client PCs.

[OliveSico]

Hi,

My worry is that PCs can potentially saturate the uplink. And i also wonder why only 4 or 5 of them on a dedicated switch  negotiate a 1Gb link. Most of NICs now support 1Gb and all are default configured (auto-nego), such as switch ports.

I have not yet spent time on uplink port counters, but will have a look asap.

Thanks

[parnassus · Accepted solution]

Double check:

• Port Settings on all "suspected" ports on each involved Switches
• NIC Settings on all "suspected" hosts (PC at 100Mbps, Printers at 10/100Mbps and so on...)
• Network Cabling

If the Auto-Negotiation parameter is set on both ends (Host side and Switch side) and all Hosts have Gigabit Ethernet capable NICs then the negotiated speed shouldn't reach only 100Mbps (Fast Ethernet) but it should be negotiated correctly at 1000Mbps Full-Duplex. If that doesn't happen on most of the examined/connected hosts you have then you should go down the rabbit hole and start to check (1) relevant newtork cabling "end-to-end", so patch cables included (Generally a good Cat. 6 UTP/STP cabling is required for properly running at 1000Mbps...probably for short runs also a good - with "good" I really mean "certified" and "tested" to work at Cat. 5E/6 - Cat. 5E. cabling should suffice for an auto-negotiated link speed of 1000Mbps) and (2) NIC Settings on each host (I presume the Switch has ports set to Auto with regard to Port Speed negotiation).

Regarding the potential saturation of the 2Gbps Uplink to a central core...well...it really depends on your network topology and usage: port statistic counters and/or a real time traffic analysis (if you have any) will tell you more to justify (or to demistify) your worries.

[OliveSico]

Ok, I setup a centreon to check 2G uplinks, ran for 2 days of workhours : my worries are gone !

1 to 5 % of Bandwidth used ...

Thanks for your explanations and time spent for me.

Cheers

[parnassus]

I'm glad you solved! Cheers.