SAN Switches

[Archipelego1]

Good afternoon,

 

if you've read some of my previous posts, you'll be aware we're currently going through the process of speccing up and deploying a virtualisation solution of 3 hosts and shared storage.

 

I'm currently working with suppliers to look at suitable switches to connect the hosts to the storage.

 

We've currently been offered Cisco SG550 running at 10GB and a HP1950 running at 10GB.

 

Although the suppliers are saying the switches should be good for the solution being proposed. I've read online that the Cisco SG550 isn't suitable due to the small buffer size, and HP themselves have said they don't recommend the 1950 for host to storage networking.

 

Could I ask, what networking are people using to connect their hosts to the underlying storage.

 

Thanks

[richbrowncardiff]

I have always looked at the specs and throughput of the SAN / NAS and then worked out what speed I need the switch to handle i.e if I'm running 4x1GBPS Nics give 120mbps max so 480mpbs max throughput. (theoretical max) realistically you will get 80-100 mbps per gb nic look at the switching (Non blocking) speed and then see if the switch copes with it.

 

If your only connecting a few hosts and san to the switch you will not exceed any decent switch's capacity.

 

I don't profess to be a switching expert before someone says I'm in the wrong but this method has served me well. For info I'm using Synology Unit with and bog standard HP 1910 Switch (£300) and configured 4 x 1Gbps nics into a trunk so the switch / NAS get 4gbps connection. This setup is used for ISCSI Luns for the File servers only. VM'S would have a 10gbps switch with trunked links.

Edited January 31, 2017 by richbrowncardiff

[mrbios]

Good afternoon,

 

if you've read some of my previous posts, you'll be aware we're currently going through the process of speccing up and deploying a virtualisation solution of 3 hosts and shared storage.

 

I'm currently working with suppliers to look at suitable switches to connect the hosts to the storage.

 

We've currently been offered Cisco SG550 running at 10GB and a HP1950 running at 10GB.

 

Although the suppliers are saying the switches should be good for the solution being proposed. I've read online that the Cisco SG550 isn't suitable due to the small buffer size, and HP themselves have said they don't recommend the 1950 for host to storage networking.

 

Could I ask, what networking are people using to connect their hosts to the underlying storage.

 

Thanks

 

Using two Cisco SG500XG-8F8T switches connected to a Nimble CS215 via 10GB here, using both the SFP+ and copper ports over 5 hosts.

 

The reason i chose those switches was due to the larger buffer size, so I think someone is giving you duff information. 16mb shared buffer is more than enough, basically so long as you have a minimum of 256kb per port (whether they state per port or shared buffer doesn't matter, if it's shared just divide it by the number of ports) then you have enough for iSCSI to be reliable. The 1950 by contrast only has 1.5mb or 3mb shared buffers depending on the size of the switch, so it's rightfully not adequate for a storage network.

 

EDIT: I'm also using some 1950's as the next step from the core to each of the 5 network segments here. They seem to do a good job for just general L2 switching at a decent price with 10GB and POE....early firmwares were buggy as hell though, seem a lot better now.

Edited January 31, 2017 by mrbios

[FN-GM]

configured 4 x 1Gbps nics into a trunk so the switch / NAS get 4gbps connection.

 

It doesn't work like that, 4 x 1gbs doesn't = 4Ggbs. The ports still run at the same speed ie 1gbs, you just have more than one connection to go down. Think of it like a motorway. When you add a lane more traffic can go down at one time. But the speed limit is still 70mph.

[richbrowncardiff]

Yes your right i didn't explain very well. you do in effect have capacity for 4gbps by trunking 4 1gbps ports though. its like teaming wan links each maybe 24mbps but if you have two you can pull down at a faster rate.

[Archipelego1]

In terms of the Cisco's. I'm looking at the data sheets and the buffers seem to show 16 megabits, not 16 megabytes?

found a couple of Redit posts that seem to confirm this.

 

Looking at the spec sheet of the HP1950, they show the buffers in megabytes (MB)

Edited February 1, 2017 by Archipelego1

[FN-GM]

It will be megabits. The Cisco models mentioned here are for small business, so I wouldn't be surprised.

[mrbios]

In terms of the Cisco's. I'm looking at the data sheets and the buffers seem to show 16 megabits, not 16 megabytes?

found a couple of Redit posts that seem to confirm this.

 

Looking at the spec sheet of the HP1950, they show the buffers in megabytes (MB)

 

Well don't i feel daft

 

Kind of cheeky how they've done that, as right above it on the spec sheet is flash and ram memory quantities in MB. However, looking over the statistics on my switches (which have been in place coming up to two years now) there's no a single issue recorded:

 

I am only using the 16 port 8F8T device with only 9 ports in use though... which only puts me one port over the limit to having 256kb per port on a shared pool. How many ports do you think you'll be using?

[njc235]

The 2920 is often used for SAN switching if you have a non converged core such as the 5900 series. The 2920 has a dynamic egress packet buffer of 6.75MB and 4.5MB ingress

 

The 1950 only has 1.5MB packet buffer.

Edited February 1, 2017 by njc235
Additional info

[Archipelego1]

I have looked at the 2920 before. But the lack of native 10GB is a pain. Yes, I can add the expansions, but it's bumping up the cost quite a bit.

[njc235]

Maybe the 2930 then which is available with SFP+ ports? Even large PBs. Packet buffer size: 12.38 MB 4.5MB Ingress/7.875MB Egres