Life after Cisco: I've got 99 problems but a switch ain't one Pending network upgrades have reignited an old debate: what exactly makes a switch "good enough?" I have the opportunity to give two switches a truly thorough battering; my lab contains a Dell PowerConnect 8132F and a Supermicro SSE-X24S. Try as I might, I can't find fault with either unit. 10GBase-T, SFP+ and Fibre Optic …
Just a warning for the future. Whilst the gear is great, the dealings I have had with Dell's router CLI made my brain hurt. It was fully capable, it just made no sense at all. I hope that they have improved that recently.
I'm glad that people are looking outside the normal Cisco / Juniper mix for routing and switching gear. I've found HP's switches to be rather good as well.
We use multicast almost exclusively for realtime control, and something we have trouble with is the time it takes for the tree to reroute when a link is broken. We've had reports of it taking 10sec or more to get going again, which is frankly terrible and tends to trip failover responses.
- We don't need anything like the bandwidth of Gigabit most of the time (only when sharing with streaming audio/IPTV), so we have no need whatsoever for 10G - however latency and jitter are seriously important.
Secondly, how long do these take to boot - both from cold and warm restarts?
Moost of our customers are not networking types (usually no IT dept at all), so being able to deploy configs using a USB stick sounds very interesting - is this available on the smaller Dells?
Just a comment regarding bandwidth requirements - very high is actually helpful keeping latency under control when you have a "glut" or batch of messages in a short period of time. This is because as soon as you get close to saturation, your messages get delayed. Even if this saturation happens once a month and only lasts a second, this may cost you serious money (e.g. financial figures announcements).
If you care about IPv6 you should do a full load test with it as some switches cheat a bit on it and offer no full hardware switching. Multicast is usually tough if you have plenty of routed ports as replication is done per subnet so for a fully portable setup (all ports with /30 subnets) would be nice to see if multicast holds.
And just to compare which Cisco switch you are throwing against?
Cisco Catalyst 4500-X ?
WS-C4500X-16SFP+ ?
WS-C4500X-32SFP+ ?
Would be nice to know how much you save going non-Cisco.
And last thing is interoperability. I know these all should support standards but some small variations might break your setup. Are you going to use these exclusively or mix with other brands? How support team would react if you call with a problem between Dell and Supermicro?
Nice to see someone sticking up for the other vendors in networking. You didn't mention hardware all that much though which would help the case for non-Cisco. In the Cisco vs. Dell battle, Dell comes off considerably better in terms of hardware and has massive and individual port buffers compared to the Cisco equivelant which is very important for bursty traffic like iSCSI to prevent flow control slowing everything down.
The interface is basically the same as Cisco on a Dell now so no problems there any more. The one area I have struggled with is all the other bits for the network such as wireless network controllers, APs etc. where it's easier to stick with all Cisco - not necessarily better, just easier. In a larger solution single vendor can be worth a lot in terms of TCO and often we find people value this more than features and best of breed etc.
You won't have problems with saturation these days unless you buy cheap switches from Ebay since all the proper manufacturers have full bandwidth backplanes so you can skip the saturation tests in future to save some time. If you're not able to run all ports at full speed full duplex at the same time then your switch is broken rather than inferior :)
unless you have one of those big fancy Ixia boxes that can flood all ports at various packet sizes and measure latency and stuff like that..
I couldn't tell quickly whether or not that Powerconnect switch is Force10 or not.. my last indirect exposure with PowerConnect (fortunately I wasn't directly responsible we had a "network engineer") the systems fell apart really quickly, at least one of them bricked themselves during a firmware update. This was about 3.5 years ago on their 6448 series. Dell support was terrible for them at the time anyway. We found out before they did that there was a newer firmware available(which didn't fix the issues in the end we went with another vendor that was not Cisco, not Juniper). I think those were OEM'd from SMC or someone. I thought it was a bad idea to go with PowerConnect(didn't mention it since both the VP and "Network engineer" seemed gung ho about them being cheap) but I didn't expect them to fail as badly as they did so quickly.
As for me, I have been Cisco free (with a few exceptions) almost my entire life. I remember I used to look up to Cisco back in the 90s and even perhaps early 00s because I didn't know any better, but since then I do.. I want to gouge my eyes with an ice pick every time I see either the IOS or JunOS interface, it doesn't have to be that complicated folks! But both companies would have a revolt after all their users have trained up if they were to go and make things simpler to manage.
Are the switches particular about the manufacturer of the SFP/SFP+ modules? I know this sounds really stupid, but the people I've asked have always told me "buy the SFP modules from the same company that made the switch".
As far as converged networks are concerned, that term is usually reserved for things like running Fibre Channel and 10 GB Ethernet over the same optical fiber (using a converged NIC, FCoE, and a switch that can either handle the FC traffic or split it out to a fibre channel switch.
Maximum number of MAC entries ?
Can it handle 1000 new MAC address per second ?
Can it handles 10000 /32 IP Routes ?
Performance under route flap - does it crash when injecting 500 routes per second for 60 minutes ?
Multicast PIM-SM and PIM BiDir is needed for VXLAN support.
How many *,G routes can it handle ?
Does it support all OSPF area types. Can you inject routes between OSPF areas at a suitable rate ?
I'm just warming up here.
MST?
RapidST?
How long until it farts and is process switching?
ASIC performance? Ports per ASIC?
SPAN ports?
Voice VLANs?
Etherchannel/LACP?
etc
Really all the article mentions is usage in a standalone access layer model, with no core or distribution layer. Hardly a real test.
Is of limited use.
The question I'd be asking is "do these switches support TRILL?" (or the other multipathing protocols)
At 10Gb per port it's pretty easy to saturate a backup link in a serious operation and given the price of the transceivers it's a helluvawaste to leave 1 merely standing-by (not to mention that all bundling protocols running on top of spanning tree have serious issues in terms of individual point-to-point bandwidth availability.)
Routing is another kettle of fish. Even Cisco don't do it particularly well in non-dedicated kit (The routing in Catalysts works, but it becomes trivial to overload the switch)
I've had a lot of problems with Cisco switches dying from bogons in the networks such as:
*MAC addresses appearing on more than one interface (once on the local interface and again on the trunk port)
*Malformed packets (bad header, same dest/source, etc)
*MAC Address/port limits on trunk ports (had this issue when we started getting more than 4000 servers per VLAN set, it would kill the core routers)
These are all issues that I've had with Cisco kit (6509's and 3750g's) and would like to see how the others handle these kinds of thing.
Found HP ProCurve switches to be very good value for money. The real ones, not the 3Com ones. The 29xx and 32xx are particularly good and have lifetime NBD warranty.
Used a pair of Dell 6648's a few years ago and suffered firmware issue, lost configs and piss poor support. Don't think they even supported Vlan priority's.
Use a 2900 as a basic L3 1gig router and have a 2910 or 2900 as top of rack with CX4 or fibre between them form a poor mans 10gig ring topology.
Anyone know about the speed of RSTP when using iSCSI and VMWare?
I'm serious - I have a customer here in Sydney with a pair of earlier Dell switches. After about 10 weeks of uptime, something goes wonky and they stop forwarding packets at or near line rate (think 20kB/s for a GigE port). So now we have a monthly switch reboot for this customer.
Being Cisco Certified, the network I am now in charge of is full of Dell Powerconnect. I SSH'ed onto a box and didn't find it that intuitive TBH. Even trying to manipulate the ARP table to find where a PC was plugged in was proving troublesome whereas it'd take me all of 10 second on Cisco. I know Dell have bought Force10 Networks and I hear good things (Especially as it's 98% cisco IOS commands) but does anyone know a decent translator where you'd put in the Cisco and get out the Dell Power Connect syntax?!?!
Why no love for Brocade? Their layer 2 stuff is hands done easier to work with than Cisco, on just about every level.
Also, how well do they implement 802.1x and MAB, are they like Brocade and will let you override a MAB password (to meet AD complexity rules), or are they like Cisco and suck at mostly everything?
For traffic flooding, try using smartbits if you can get your hands on the kit, at tradeshows they claim to be able to generate traffic to find a switches breaking point, could be a useful metric.
Would be interested to see how standards compliant layer 3 switches are, do they interop with Netgear, Dlink or Allied Telesyn, even each other in terms of vlans etc
One flaw in smart switches that we saw about 10 years ago in the data center with an SMC l3 switch, and more recently on a client site with Netgear L3 switches is something I suspect that has occured for the Australian who commented above. In the lab, the switches will run forever without a problem, but in the field, they lock up periodically, the SMC used to lock up about once a month, and the Netgears about once a week, but both run fine in the lab. Both scenarios took an internet feed over cable into the switches. Once you take that internet feed out, put it into a 5 port dumb switch, and then link the dumb switch to the smart one, the lockups do not re-occur. This to my mind is a good reason to run with Cisco etc in this applications. Never investigated the problem further, but assume it is something like an ARP cache filling up to breaking point when hosts on the switch are being bombarded by probes from script kiddies. Replicating that in the lab and determining susceptibility would be useful, then you can class the switch as suitable for front line use or for behind a firewall only use.
Try dropping in a new $500 switch and see how it performs compared to the other two. If you can't tell the difference with that load, you are not going to be abel to test the differences between these two more expensive switches. In my tests, it looks like the cheap generic switches are just as good as the ones that costs 20 times as much.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
