Switch Speed and Price – Tradeoffs
March 15, 2010
How fast should your switches be?
Sure, the answer is “as fast as you can get”, but we don’t all have the budget for, say, this beast (chassis sold separately). Lots of us don’t have money for even a 48 port Gb managed switch.
So when deciding on a switch to buy, there are a lot of variables to examine.
First, the hard requirement is the number of nodes we have to connect to the network. Then we have niceties such as management (so that we can configure and monitor them, use VLANs, etc etc), whether or not they’re stackable and all that…but where does speed come in?
Obviously, a 100Mb/s switch costs less than an otherwise equivalent Gb/s switch.
The rule of thumb I’ve been operating by is that the end user switch shouldn’t be the bottleneck. In other words, the users in my offices talk to servers over the WAN. I’m never (not soon, anyway) going to get a GbE WAN link between my sites. Because of that, the 100Mb/s switches we’re using don’t hold us back. We don’t share files between the users directly, so anything bigger would be overkill.
If my users used local servers, though, an increase in switch speed would dramatically improve the response of the server. The quality of experience would improve, and it would be worth upgrading the switch.
What kind of mental calculations do you use when picking a switch? I’m interested in learning if there’s a “right” way, and if not, maybe we can aggregate all the ways we make the decision into some smart ideas…comment below!
Posted in General
Email me
content rss
March 15th, 2010 at 11:40 am
I’m very old-school. I still advise people that for 95% of the work loads out there, 10/100 to the desktop is fine (ie one 48-port switch with one or two Gb uplinks). This assumes that the workload is web, word-processing and associated activities, and all important data is stored on central storage (Rule: if it isn’t on the filer, it isn’t backed up!). Yes there are exceptions, but by making them exceptions and making them prove their requirements you can cut down a lot of cost right there.
What happens beyond there depends on the complexity of the network, the number of computers involved, the job being done, and (let’s be honest here) the budget available.
March 15th, 2010 at 12:13 pm
I work on the data center side so everything is GigE or higher, but I would not see why you do not want GigE on the end-user side in an office. We have GigE and wireless in our office. Why? Because the cost differential between 100MB and GigE was trivial for the small 16 port unmanaged switch we require. Now if I had to buy a 100 of them that may be different.
Several trends make me over-deploy on the network
- Online Backups: Increased used of off-site backup services (Mozy, Carbonite, Symantec and others)
- SaaS: Increased used of Software as a Service. Salesforce, Freshbooks, Google Apps and similar tools mean that more and more desktop functions now require bandwidth
- Rich Media: Email, web conferencing, web sites, and more are increasingly using more bandwidth.
- Faster Access: Comcast just deployed 50/10 Mbit service in my area. As speeds increase, more and more functions can be done over the network. For example, we could switch to VoIP and have plenty of bandwidth for other activities, especially if a managed switched was used.
March 15th, 2010 at 12:16 pm
First, beside the speed of the single port, the total bandwith of the system will be important to me, i.e. how many clients can communicate at the maximum speed before the fabric speed will be the bottleneck. What is needed here is of course highly dependent on the specific case.
Next, I care if I can link more than one switch into a bigger one for easier maintenance/management and how fast this uplink will be.
Also, I reassign ports into different VLANs relatively often. Is a web interface the only config option or can I automate this, via a CLI or SNMP? This is important for general management as well if you have many switches. It’s hard to integrate a web config interface into any NMS.
For me, security is a growing concern: Is SNMPv3 offered, or at least SNMPv1/v2 limited to specific ports or addresses? How about port security? 802.1X auth?
March 15th, 2010 at 10:06 pm
In my situation, we recently upgraded the network from a flat design(we still had hubs 3 years ago, I’m serious) to a broken up design with subnets. It was a large scale upgrade that we were lucky to bury in a capital project. Even still, we couldn’t go crazy. Since we already had 30 Cisco switches and a Cisco VoIP system already deployed at two of our schools, we decided to stay consistent and build off of that. Beyond that, we have waded into the pool that is managed wireless and stuck with Cisco there as well. We needed POE switches for both the phones and the access points.
We ended up going with Catalyst 4507R core switches in two buildings to go along with the existing 4506 that serviced the school with the Cisco phone system. From there, each wiring closet had a Catalyst Express 500G as a distribution switch connected via a port channel(sort of like NIC teaming) back to the core. Each access level switch(CE500) connected to it’s 500G via port channels as well. Administration of the CE switches is somewhat limted compared to it’s fully featured Catalyst cousins, but they have performed well. All routing is done at the core switches, which double at access level switches as each has a 48 port POE blade.
The classic Core>Distribution>Access design is blurred at my org because of available resources. Basically, the backbone all of the way to the access level switches is at least Gig speed. Server uplinks to the main core switch(es) are gig. All workstations are 100 Meg.
March 17th, 2010 at 7:28 pm
I tend to agree with the sentiments re: not bothering to spend money when the bottleneck is clearly identified. If I’ve got a location w/ no server computers and a T1 MPLS “pipe” back to the hub site gigabit Ethernet switches aren’t going to make any difference. If I *EVER* see any of that 50Mb Internet service I’ll change my tune, but in my area of the midwest we’re happy w/ our “prosumer” grade cable-based Internet access and our entire
In a network that has any amount of complexity, though, monitoring is essential and “mental calcuations” should go right out the door. Without monitoring you only have your “gut” to tell you where your bottlenecks are and whether or not spending bigger money on speed, fabric bandwidth, etc, is really worth it. Cacti, MRTG, and similar tools are no cost and easy to deploy. I actively worry when I can’t see long term traffic trends and short term bandwidth utilization. Not knowing what’s going on inside the wires gives me one less method for capacity planning and one gigantic blind-spot when troubleshooting anomalous network behaviour.
I concur re: 10/100 being sufficient for many desktop applications, and I’m not seeing a push to pull out existing working 10/100 infrastructure in most “task oriented” work environments. A lot of that gear is aging, but if you’ve got a good pool of spares and the bandwidth is what you need don’t bother replacing it. I’m seeing new gigabit gear sneaking into these otherwise 10/100 networks in the form of secondary networks to support backup or replication systems, and sometimes in tiered distibution models to drive more client bandwidth, aggregate, to server computers.
If I’m going in “green field”, though, I’d be hard pressed not to do 10/100/1000 unless the Customer was *really* strapped for budget. Gigabit is just too cheap today (relatively speaking).
March 18th, 2010 at 12:14 am
While I appreciate the sentiment, and personally keep much old equipment going for a ridiculously long time, in there it is implicit that you started with something good to begin with… Gigabit has been around for so long that prices are about as low as you can expect them to get. I would hold off it was new and had a steep premium, but at this point, I’d call gigabit a prereq.
Through multiple companies, some very large, some very small, the lesson has been hammered into my head in some very unpleasant ways, that squeezing through the low-end will bite you, and in short order… How low is too low? Why not get a 10Mbps switch? It’s faster than you need! 1Mbps? You’ll soon find yourself in an equivalent position. I’ve seen plenty of shops where they were they aimed low to save a few bucks, which has resulted in the whole place stagnating… Oops, can’t use this app, or that app, PCs are so old it won’t work. They take 5 minutes to boot-up now, because they were old when they got them, and now the software has been upgraded, but not the hardware to save money, and they’ve become painful. But hey, you can’t ever get them upgraded, because they were good enough years ago, and they’re still working. Employee time is cheaper than an upgrade…
Consider that if you don’t upgrade now (when you have a need to fill) you may not, for decades to come. Sure, 100MBps is okay, but it’ll be pretty damn creaky in just a few years here… Network requirements always shoot upwards. VoIP? Network-based backups? System cloning? etc. Not to mention that BUYING ultra-low-end misses the point… You might keep using a gigabit backbone switch in other contexts for decades to come, but 100Mbps? You’ll probably just toss it as soon as you can finally get an upgrade… In fact you could probably find 100Mbps managed switches in the trash of other companies. The price difference isn’t huge, while the performance difference is… always the sign of a good place in the bell curve to be at.
March 18th, 2010 at 3:31 pm
I think it’s funny that you have a link to switch on Best Buy’s website. If this is where you are getting you networking equipment…. I need to ask… How long have you been in the tech field? Shop at really tech stores like CDW, GHA Technologies, or B2B Computer Products.
My only recommendation for switches is HP ProCurve and their “Lifetime Warranty”. The switches are more expensive than your Dell, NetGear, Linksys… etc… but the reliability of the product will make you happy your network will not go down in the middle of the day… causing you aggravation and users wanting to kill you for not preventing the outage…. that being said.. you should always have spare switch equipment in your arousal…
March 18th, 2010 at 3:55 pm
Funny Brian.
No, I just needed a link to a generic 48 port managed Gb switch.
Despite the fact that we’d all really like enterprise-quality hardware, it’s not reality for everyone.
Besides, quality hardware would take all the unpredictability out of life
March 23rd, 2010 at 1:30 pm
Matt,
Have you looked into the pre-owned market? I get several emails each week from vendors looking to buy or sell excess equipment. They give them a thorough lookover prior to sale and also offer extended warranties. I know several people that buy through these kinds of companies. I bought some Cisco waps from one of these vendors a couple of years ago, and they haven’t failed yet. Some items can be had at a pretty steep discount. Or, you can buy refurbs from Cisco directly. I can send you some links of you like.
Curtis
March 23rd, 2010 at 1:35 pm
Hi Curtis,
Thanks, and yeah, I’m a big fan of buying refurbished network gear. I don’t know if I’d say the same for servers themselves, but I’ve almost never had a problem with the reliability of the hardware I’ve gotten.
I typically use Network Hardware Resale, and they’ve been pretty solid.
You really hit on the important parts of buying refurb. Warranty, reliability, and price. If those three are there, it’s like buying retail, but cheaper
Thanks for the comment, and I agree. As for your offer, I’m absolutely interested in hearing who you choose. Drop me a line on here or at standalone.sysadmin@gmail.com
Thanks!