So whats everyone using for switching and servers?

[kevin_hsieh]

My servers are connected to Cisco 4948 or 4948e switches. Anything less than the 4500 series from Cisco really isn't recommended as they don't have the port buffers required to handle microbursts, and you will see lots of dropped frames. I was losing over 1-1.6% with EqualLogic on 3650E switches. My CS300 is connected to some 10G ports on my 6509. All my switches are used, which saves me a bundle. Currently running on R710, which I plan to start replacing in 2016 with newer gear and 10G networking on the host.

[iannoble38]

Cisco 9372TX in NX-OS mode, 48 ports of 10gb base-t & 6 x SFP 40gb uplinks, good value as it was the first Nexus kit in our environment.

I was originally looking at the trident II based switches, the 9372 is trident II too but it also has a Cisco custom asic chip - I liked the extra buffers & micro burst protection that this added.

Could have went for the SFP based model but the extra up front and increased smartnet expense wasn't worth it for our environment, the Nimble supports SFP or 10g Base-T.

[mtedross96]

We are looking at upgrading our Dell power connect 1G iSCSI switches to an affordable 10G-BaseT switch.

We were leaning towards the Netgears. What issues did you have? were you able to fix them through Nimble? or Netgear's support?

I am have a CS300. do you happen to have the same Nimble SAN model?

[pbitpro96]

I would stay away from NetGear... just not worth the perceived value!

For a reliable, solid performing, "value" switch, I've used teh Brocade ICX 6610 with good results for small SANs

Brocade ICX 6610 Switch - Brocade

[dejvnull49]

Hi!

Update on the Netgears... they were very very buggy when trying to add L3 support to them... and really don't use the stacking feature it just is not worth it.

If you run then as L2 switches and have them really dumbed down then they might be ok. Their support is a non-support, overall pretty much non-existent.

After having to many issues with Netgear we switched over to Arista 10GB, which are totally the opposite of Netgear, they are great, their support is great.

We now took the Netgear back in to production for a different set of servers, we run them redundant and non stacked, dumbed down. The only thing we set in them is some VLAN's and Jumbo frames.

Things we know have been very problematic previously are L3, Stacking, NTP, and the web interface.

L3 was just really broken...

All this might be fixed in later updates of the fiwmware, I don't know.

So to finalize, I would not get these for mission critical systems. There are other vendors in a similar price range that are much better. If you can afford then go with Arista they are the shit, and really really stable!

/Dave

[jayst136]

i'm looking at Netgear 10Gbit switches as well. Specs seem good enough regarding buffers and latency, all at a very reasonable price.

M7100 | Fully Managed Switches | Switches | Business | NETGEAR

I would only use them for L2 iSCSI traffic. Would not put them in core or let them provide any L3 services.

I'm very happy with Dell R630 and R730 servers ath the moment.

[lindy37]

Cisco 5548's for LAN and iSCSI

Dell R630's for VMware 6.x hosts.  Each host has 4 - 10gig ports (Intel X520-2?) 2 for LAN, 2 for iSCSI.  We run Jumbo Frames on the hosts/vmware and switches (preferred send/receive).  Nimble CS300 10gig, Jumbo on.

[chris24]

It really doesn't matter what switches they are providing they adhere to one rule - they have sufficient buffer cache to handle the throughput. You will know when it's an issue as a performance test will result in a saw blade graph where the cache becomes saturated. Re-transmits and high pause frame count are also indications of this saturation.

Aim for =>512kb per port, most vendors however share one pool across all the ports! It is for this reason that stacking switches is bad news as any inter switch link (ISL) traffic will consume this cache also, this highlights the importance of ensuring you isolate your paths via different subnets or event & odd or bisect.

All switch vendors publish these stats, if they don't - don't buy the switch.