Best 10GbE Switch on a Budget?

KapsZ28

2[H]4U
Joined
May 29, 2009
Messages
2,114
We are finally upgrading our ESXi hosts and NetApp to 10GbE but are limited to how much we can spend and get approved. I would like to get at lest 24 ports, stackable switches with SFP+ ports. Trying not to go over about $10,000 per switch. One of the switches we were looking at is the Brocade VDX 6740, but I would like to know what else is out there for the price that may be better than going with Brocade.

I am curious about Extreme Networks switches, but with the lack of online pricing I am guessing they are very expensive.

I was also reading this article yesterday, http://www.tomsitpro.com/articles/dell-s6000-s4810-switches-cumulus-linux,1-1628.html where they take the OS out of the hardware which sounds great. I know VMware is looking to do the same thing and these where designed by engineers from Cisco and VMware. Only problem is they start at $23,000 per switch. Then again, it is twice as many ports as the Brocade.

Another thought was going to Blade Servers. We wanted to get two more servers budgeted at about $8,000. Right now everything we have is whitebox, and I would prefer to go with Dell. We just so happen to have a Dell PowerEdge M1000e blade enclosure. The servers and switches are garbage, but they can all be replaced. Only pricing online is for the blade servers, not switches. We would need two 1GbE and two 10GbE switches for the blade enclosure. I contacted Dell, and am waiting for a rep to get back to me.

Any suggestions?
 
If you want to have serious DC solution that you will think about Nexus 3k or 5k it depends on what kind of features you need. If you need SDN prepared DC network then look at Nexus 9k.
If you will have blade centre think about Cisco UCS. Check UCS you will not be disappointed.
 
SDN would be great if it was in budget. Also the Cisco Nexus 5548p which is a 32 port 10GbE switch is $17k. Well above the $10k per switch budget.
 
You don't buy a 6509 Catalyst to do 10Gb. The price per port is way too high.

The Nexus 3K series is just a bit above your budget...but also check out the new Meraki 10Gb switches.
 
You don't buy a 6509 Catalyst to do 10Gb. The price per port is way too high.

The Nexus 3K series is just a bit above your budget...but also check out the new Meraki 10Gb switches.

Yeah, I have worked with the Meraki WAPs. Just wasn't really sure how good they are. Is Dell garbage? I'd be interested in a Dell S4810 since it is 48 ports.
 
Force10 is not garbage by any means. I have yet to run into a single problem or performance issue. One of my customers last year, big name customer (well known brand) went full Force10 for their CORE. M1000e's with Aggregators, no problems..they love the solution.

Why do you need 1Gb along side 10Gb for your Blade Fabric?

You could get by with a couple of Dell M I/O Aggregators in the M1000e, and leverage NIOC within vSphere dvSwitch.

Uplink to a couple of Force10 S4810 ToR. Collapse your Storage traffic into the Force 10's.
 
Force10 is not garbage by any means. I have yet to run into a single problem or performance issue. One of my customers last year, big name customer (well known brand) went full Force10 for their CORE. M1000e's with Aggregators, no problems..they love the solution.

Why do you need 1Gb along side 10Gb for your Blade Fabric?

You could get by with a couple of Dell M I/O Aggregators in the M1000e, and leverage NIOC within vSphere dvSwitch.

Uplink to a couple of Force10 S4810 ToR. Collapse your Storage traffic into the Force 10's.

The only real reason for the 1GbE was for management. I guess it isn't really necessary as it can all be run through the 10GbE connection.

I hate to ask, but what exactly are aggregator switches?
 
NetGear XSM7224S

You don't say what features the switches require but that's 10GbE stackable and under 10k
 

Yeah, I also watched the video. So it automatically makes LAG groups and simplifies other stuff. I like the management from the CMC. I guess the question I am asking is why would I choose the I/O Aggregator over say the PowerConnect M8024. Are there any downsides to going with the aggregator? If we went with it, would you get the blade servers with two or four port 10GbE NICs? And any idea how much they cost per switch?
 
Personally I like the aggregator for larger deployments or where ToR makes sense and because it's truley converged, easy, and affordable. It's also an investment and supports all your storage protocols. I've also used the 8024's but more so on smaller deployments and collapse storage right into the blade fabric, plus you get layer 3 capabilities as well with the 8024's.

As for blade fabric you'll want two for redundancy, A Fabric and just leverage the 10Gb LOM on the blades.

Cost is reasonable I rarely get into that conversation that's usually Sales.
 
Hopefully I hear from a Dell rep tomorrow and start getting some pricing.
 
We're running ours on a stack of Dell 8132Fs.

I'd used force 10 if we could have afforded it.
 
We're running ours on a stack of Dell 8132Fs.

I'd used force 10 if we could have afforded it.

How is the 8132F? They are within budget for us. Still waiting on a Dell rep, but I doubt we will get any discounts since we haven't been using Dell.
 
They're ok, but we arn't doing anything special with them.

We had (6) 6x1GB lags running through our Alcatel Stack of 6850s.

We redesigned the network for 10Gb and purchased 3 8132Fs for the dev server room.

We went from running 7 Alcatel 48 port switches at 96% CPU for hours on end to never swinging above 25%.


The Dell 8132s never see more than 15%. Buying (3) 10G switches was overkill but chewing up POE ports on the other switches cost almost as much.

We have (2) 8132 stacked and (1) in another room. They work fine for what we do, but we aren't doing anything special. 100 TB of transfers a month mostly, but it comes in 2-8TB bursts.
 
mention who else you are considering and what they are offering you. Dell will often price match or beat a competitor to get you to buy their switches.

Another thing of note: the 8132 is fairly quiet for a L3 10Gb managed switch.
 
Last edited:
mention who else you are considering and what they are offering you. Dell will often price match or beat a competitor to get you to buy their switches.

Another thing of note: the 8132 is fairly quite for a L3 10Gb managed switch.

Will give it a shot if a Dell rep ever bothers to contact me.
 
Will give it a shot if a Dell rep ever bothers to contact me.

If you have an issue getting assistance, let me know...shoot me a PM and I'll take it up the Channel.
 
So I am reading Gartner and they are also talking about SDN being a key component in cloud computing. Being that cloud computing is exactly what we are doing, are there any 10GbE switches out there that are SDN an say no more than $12k a piece?
 
So I am reading Gartner and they are also talking about SDN being a key component in cloud computing. Being that cloud computing is exactly what we are doing, are there any 10GbE switches out there that are SDN an say no more than $12k a piece?

give supermicro switches a look. SSE-X3348S does SDN i believe and is around 12K
 
So I am reading Gartner and they are also talking about SDN being a key component in cloud computing. Being that cloud computing is exactly what we are doing, are there any 10GbE switches out there that are SDN an say no more than $12k a piece?

Are you a cloud computing provider? That's really where SDN's big play is at the current time. It makes a lot of sense for cloud providers because they need to be able to change their offering quickly, adapt to the trends, etc. Much easier to do when it's all in software vs. needing to do major physical changes.

SDN also makes a lot of sense in larger enterprise networks that have a lot of change going on.

Otherwise, you would probably find that adoption rate for SDN is "cautious" because it's still very new and has higher cost of entry. Small and mid size networks aren't usually changing much, so the benefits are minimized.

There's also the fact that SDN and SDS(torage) are things becoming baked right into hypervisors with every new release and when it comes to performance, it can't get any better than this stuff being native in the hypervisor code itself, so a lot of people want to see how that matures before jumping on those bandwagons.
 
Are you a cloud computing provider? That's really where SDN's big play is at the current time. It makes a lot of sense for cloud providers because they need to be able to change their offering quickly, adapt to the trends, etc. Much easier to do when it's all in software vs. needing to do major physical changes.

SDN also makes a lot of sense in larger enterprise networks that have a lot of change going on.

Otherwise, you would probably find that adoption rate for SDN is "cautious" because it's still very new and has higher cost of entry. Small and mid size networks aren't usually changing much, so the benefits are minimized.

There's also the fact that SDN and SDS(torage) are things becoming baked right into hypervisors with every new release and when it comes to performance, it can't get any better than this stuff being native in the hypervisor code itself, so a lot of people want to see how that matures before jumping on those bandwagons.

Yes, our two main markets is networking, we are an ISP with peering sites all over the country, and we are cloud computing. Other stuff as well, but almost all my focus is cloud computing.

So, is the Brocade VDX 6740 considered SDN? I believe that is just a buzz word, but I wanted to make sure it was software based.

I heard networking is eventually going to be built into VMware which would be awesome!
 
So I want to go with two dual 10GbE NICs per server with two 10GbE switches. Keeping the production and storage VLANs separate. My CTO doesn't want to. He wants just a single dual port NIC and have both switches and ports configured identical with overlapping storage and production VLANs. Is it a bad idea to share a 10GbE port for both production and storage? We are only using NFS.
 
They say it's best practice but if you only have 2 links and you split them then you have no fault tolerance. I would say split the traffic by VLAN but let them use both links in that case. If you have 4 links then I would split them like this:

Card 1 Port 1: Users
Card 1 Port 2: Storage
Card 2 Port 1: Users
Card 2 Port 2: Storage

With Card 1 going to SW1 and Card 2 going to SW2.

In no situation would I suggest sharing the VLAN but sharing the port...meh, depends on the load.
 
Yes, our two main markets is networking, we are an ISP with peering sites all over the country, and we are cloud computing. Other stuff as well, but almost all my focus is cloud computing.

So, is the Brocade VDX 6740 considered SDN? I believe that is just a buzz word, but I wanted to make sure it was software based.

I heard networking is eventually going to be built into VMware which would be awesome!

I believe Brocade Virtual ADX and Vyatta router are the actual SDN software components. You still have the physical interfaces (switches) that are needed. Some companies have specific switches to work with their SDN software and some companies don't and will work with "anything". Not sure what Brocade does here on the physical side to complement SDN.

VMware purchased Nicira to get into SDN and they now market it as NSX and I believe it's in GA release.
 
So I want to go with two dual 10GbE NICs per server with two 10GbE switches. Keeping the production and storage VLANs separate. My CTO doesn't want to. He wants just a single dual port NIC and have both switches and ports configured identical with overlapping storage and production VLANs. Is it a bad idea to share a 10GbE port for both production and storage? We are only using NFS.

You can get away with storage and "other" traffic, however i would recommend that you use some sort of QoS like Network I/O Control within vSphere Distributed Switch. From the previous post if you're still talking about the Dell 1000e's then that's your option. This is one area where Cisco UCS is king...because UCS does QoS in hardware.

VLANs are for logical segmentation, nothing more. VLANs aren't going to help you when you're storage starts taking bandwidth from Virtual machine traffic, or if you evacuate a host for maintenance and all the vMotion traffic is having a negative impact in the same fashion.
 
Last edited:
So I want to go with two dual 10GbE NICs per server with two 10GbE switches. Keeping the production and storage VLANs separate. My CTO doesn't want to. He wants just a single dual port NIC and have both switches and ports configured identical with overlapping storage and production VLANs. Is it a bad idea to share a 10GbE port for both production and storage? We are only using NFS.

Either route will get the job done. I would look at it from a throughput need standpoint and how you planned to load balance or team. Keep in mind load balancing of NFS requires multiple IP targets (just load balance teaming NIC's doesn't get it done).

You have potential for 40GbE of throughput vs 20GbE, will that make a difference and justify the extra cost of NIC's and switch ports?

You will also want to make sure traffic on the port is prioritized appropriately
 
Yes, our two main markets is networking, we are an ISP with peering sites all over the country, and we are cloud computing. Other stuff as well, but almost all my focus is cloud computing.

So, is the Brocade VDX 6740 considered SDN? I believe that is just a buzz word, but I wanted to make sure it was software based.

I heard networking is eventually going to be built into VMware which would be awesome!



the VDX is pricey due to their software they run, logical chassis stuff(datacenter cloud-stuff), trill for l3 over l2. we have them and they are great, but they cost a TON

unless you are doing a large deployment of them, the ICX makes more sense since you wont use the $$$ features of the VDX

we use hp pro curves for some 10 gig as well, the 5406 chassis, they take up some space, and will over subscribe, but $ per port is great, atleast with out pricing, I haven't really looked at the COMware line of stuff from HP

the juniper ex4550 is a great packet pusher as well
 
I am also using Force10 at the CORE of a large institution.

Our SAN uses a combination of PowerConnect 8024 and 8164. The powerconnects are good for basic 10Gb L2, I wouldn't use them for anything more than that personally. The CLI needs a bit of work.

One thing I do like about the PoweConnects, you can stack switches within the same model line. For Instance, if you needed a combination of SFP+ and 10Gbase-T, you can stack a 8132 + 8132F, or even a 8164 + 8132F.
 
oh yeah, ARISTA!

they seem big into SDN, and they are usually great price wise supposedly
 
Back
Top