Sign in with
Sign up | Sign in
Your question

Best gigabit switch

Last response: in Networking
Share
April 3, 2006 4:26:09 AM

Which switch do you you guys recomend for a small home network (I have a 5 port gigabit linksys one at the mo but it seems to fail on giving me decent speeds)

Also out of the big boys, who makes the fastest 24 / 48 port 100 mbit switches?

More about : gigabit switch

April 3, 2006 2:14:43 PM

Since I'm Cisco"centric"... I'm going to say Cisco.
April 3, 2006 7:16:46 PM

Cisco is pretty much the hands down god of the networking industry so you will be guaranteed good results from their equipment, if configured correctly of course ;) 
You had better have a deep pockets though...
Related resources
April 4, 2006 1:48:01 PM

For the 100Mbit 24/48 port switches, a good alternative is HP. Lifetime warranty on their products, nice interface, and you don't need to keep paying Cisco for service maintenance contracts.
April 4, 2006 10:19:46 PM

thought HP ran the Cisco IOS on most of its top range ones?

i have catalyst switches here (24 and 48 port 10/100s with 2 gigabit ports) however the gigabit never usualy works that "fast" -- any recomendation on the NICS that work best wtih them ?
April 4, 2006 10:34:22 PM

You know... Ethernet throughputs, I think are pretty much tapped out.

I honestly don't think there's too much difference between any of the non-blocking switches out. This is especially true if you're talking about the stand-along 12-48 port versions (+ or - a couple of gig ports) out there.

So I say this being a "cisco" guy, it doesn't really matter... HP Procurve, Nortel, 3Com, Foundry, even some of the higher-end NetGear stuff... they are all good for the most part.

The real differences come in what features you get for the money (802.3ad Port Channels, 802.1q/p Trunks, QoS features, stacking, etc.). Also, the bigger multi-slot chasses (Cisco 4000's and 6500) are a different story, each model of line card comes with different backplane bandwiths, etc.

As to NIC's... there's not too much difference anymore between 100Mb NIC's. For 1000Mb NIC's... Look for ones that offload the most off the CPU. The bottleneck right now for PC GigE NIC's is I/O & CPU bottlenecks. So anything that takes load off the PC's will gain you a bit of throughput.
April 5, 2006 12:02:05 PM

Fair enough, It's going to be used for a large LAN so I guess I will go for a newer cisco one (which I can pillage from the office) -- I have used a 2950 before for a LAN however and the max speed we could obtian over it was about 4mb/s from one PC to another (via full duplex 10/100/1000 nics -- set at 100FD into 100FD ports on the switch) When I asked here the reasons seemed to be the differenet network cards in each PC so I was wondering if it matters which cards they are, or if its just they have to be the same card...
April 5, 2006 4:05:10 PM

Quote:
I have used a 2950 before for a LAN however and the max speed we could obtian over it was about 4mb/s from one PC to another (via full duplex 10/100/1000 nics -- set at 100FD into 100FD ports on the switch) When I asked here the reasons seemed to be the differenet network cards in each PC so I was wondering if it matters which cards they are, or if its just they have to be the same card...


Really? 4Mbps?! That's pathetic. Something is fishy there. Connect the two PC's with a crossover and see if you still get only 4mbps. That could help you rule out the switch or not.

However, I have seen a 2950 push 100% of a 100 and 1000port in the past, so I know they can do it.
April 5, 2006 4:21:20 PM

A 2950 should get way more than 4mb/s, either the nics you were using were really crappy or it was having major problems.

For home give a buisness line netgear a try. I have a 16port gigabit netgear and I average 40megabytes per second when transfering large files from my machine to my file server. I'm sure the servers hard drive speed is the only thing slowing it down.

For enterprise use go for a cisco if you have a buttload of money, otherwise get an HP. If you arn't going to use all of the extra goodies in IOS you might as well just get an HP at 1/10th the price.

Using different network cards should not have a huge impact in network performance. Using really crappy/old network cards might. I myself prefer higher end 3com or intel network cards.
April 5, 2006 6:05:44 PM

see the odd thing was that the PCs were mostly using onboard network cards(most were intel ones) and none of them were of bad spec. However with cross over cables, the transfer speeds still were quite bad. However we had no idea why :( . the pcs were all runnin XP with SP2 etc, no firewalls (we tried them on // off with no noticable difference)

So i guess it must have been the pcs or network card settings or somthing, just totaly unsure why :S

(all Network ports everwhere were set to 100FD or gigFD where they could be)
April 5, 2006 6:21:38 PM

Maybe the machines got a bad set of drivers, I've had some issues in the past using drivers windows finds instead of installing them myself.
Could always be an issue with another piece of hardware in the machine as well. I've had hard drives get kicked down to PIO 1 mode before. Dont remember if it was a bad cable or bad motherboard but one of those did it.
April 5, 2006 6:52:27 PM

:S they were all running the latest drives as far as I know (ms update usualy takes care of those things now) Its just really odd :S
April 5, 2006 7:13:11 PM

...that makes me wonder if there were IRQs being shared.

The again, we are talking a out XP - so all sorts could have been going on.
April 5, 2006 10:35:59 PM

best speeds i have ever got was 8mb/s from gentoo to gentoo on identical boxen, on windows best i have seen was 6mb/s. never anything better, not even gigE to gigE
April 10, 2006 2:26:10 AM

I would say Cisco is the BEST hands down. Although, I work for a technology department at a school district, and we use Nortel, and they seem to work fine. We get about 6-7MB/sec throughput. As far as the load on the switch, that is with all 48 computers receiving a Norton Ghost image through the 48 port switch. So that is pretty much full throughput. No slowdowns no matter if 1 computer or all 48 are downloading the image from the server. We have 48 port gigabit switches throughout the distict. The 6-7MB/sec is obviously on a 10/100Mb ethernet card. I haven't tested gigabit speeds yet.

My System:
http://amdgamingrig.dyndns.org
April 10, 2006 8:48:19 PM

And what 24port/Gigabit switch would you suggest for a Netcafe/LANparty place i m helping a friend to make?

I found the Linksys 24 port unmanaged, is Linksys as good as ppl say (due to Cisco family relationships) or would you guys recommend me another model?
Pls note that Cisco is double the money of the Linksys!
April 11, 2006 6:26:51 AM

I assume you are going to be connecting the 50 computers you mentioned in the other tread?
Which linksys were you looking at exactly? Is it the SR2024?
If you cascade a few of those to get 50 ports it will run into some serious bucks. Using the fiber SX modules to uplink the switches would be a must, those alone are probably over $200 apiece.

If you want to save some serious bucks I'd consider going to 10/100 switches instead of gigabit. If all you are going to do on the computers is gaming/web surfing you will not see a difference between 10/100 and gigabit. You could then buy a high end managed HP 2650 that has 48 10/100 ports and 2 ethernet/fiber gigabit ports for $800 and have everything you need. If you run some dedicated servers just put them on the two gigabit ports and you will really be set. 50 computers could not possibly generate enough gaming traffic to make that thing hurt. Also, it is only 1U so it wont take up a lot of space :) 
I have quite a few of those in service at work as backbone switches for large schools.
April 11, 2006 8:04:54 AM

yes it's the SR2024 that has optical links on it
This is the model

the cost approx 400USD each, and I ll need two of those and one 16 port since I m having 45-50Clients. Can't this two Linksys connect between themselves with the optical link? Or do i need an optical switch to connect them?
I need Gigabit LAN for sure, all PC's have onboard Gigabit (Nforce4) since many games will run from the File Server (older games images iso's and also save/load game server)
April 11, 2006 2:14:24 PM

As a network engineer, I never recommend cascading switches. If you're going to need that many ports, it's always better to get a modular switch. Connectivity between the two switch is dependent on that etherchannel and you're always limited on that bandwidth between the two. Plus, you could also screw up spanning tree, becomes more difficult to manage and theres limited scalability.
April 11, 2006 4:52:05 PM

You could cascade the linksys switches but they do not include the fiber modules, they are extra. After buying two of those switches and a fiber module for each (and possibly another for another switch) you are looking at over $1500 in equipment. And like el0him said, cascading isn't to great of an idea.

If you can afford it the absolute best solution would be an HP 4100 series chassis with some 20 port modules in it. The 3U 4100 chassis would hold about 80 gigabit ports, the 5U would hold 160. An HP 4100 chassis with 3 20 port gigabit modules (60 ports) will probably run around $5000.

Otherwise I think the HP 2650 would still be your best bet. If you did end up needing to cascade another one later it has gigabit copper ports right on it so you dont need to buy a fiber module. Since it is a managed switch it wouldn't be quite as terrible to cascade a switch off of it.
April 11, 2006 5:19:37 PM

ok well now i just did gige -> gige over a linksys 5 port switch and i got 8mb/s i mean WTF!? I give up on gigabit stuff
April 11, 2006 6:45:57 PM

The llimitations you are facing are probalby from the computers ability to process the packets, try to see if you can increase the input buffer on the network cards. Also remember that packet size also determines how efficient you are. Small packets means less overall data throughput. Remember, number of packets != amount of real application data. You have all the overheads to account for.

Quote:
ok well now i just did gige -> gige over a linksys 5 port switch and i got 8mb/s i mean WTF!? I give up on gigabit stuff
April 11, 2006 9:55:07 PM

i can't find the HP's here in Greece guys, either the Linksys 48 port Gigabit switch, so a mate that works on networks said to get 2 pieces of those 24port linksys and patch them using 2-3 patch cord cables, so i get 3 channels to prevent bottlenecks. Is that correct or wrong thinking ?
April 12, 2006 1:49:51 AM

You can't even order an HP Procurve over there? That sucks :( 
HP or CDW should ship internationally. After paying $1000 or more for a switch a hundred or so for shipping ain't to bad ;) 

Linking the switches together using multiple patch cables wont remove bottlenecks, it will probably take down the whole network and possibly damage the switches. If you try to connect up all of the switches with ethernet patch cables you will most definatly run into problems when the swiches are under load. Getting as many ports as possible in one switch is your top priority.
April 12, 2006 4:16:28 PM

Don't know much about those switches you are trying to buy but you will most likely run into spanning tree problems if you have multiple ports connected between the switches.
April 13, 2006 12:59:28 AM

Well I the gentoo server up using a liveCD (gentoo as well) and managed to get 20mb/s both ways which is better, but still crap.
April 18, 2006 7:44:03 PM

OK, i got the Linksys Gigabit switch and i did some testing today with the help of NetMeter Beta v0.99.

On one side i got my IBM xSeries 226 server with a Broadcom gigabit ethernet running Win2k3 Server and on the test PC a onboard Asus Nvidia NF4 Gigabit ethernet. Both machines use good Cat5e cable (Belkin)

Copying a DVD image (big file ~5gigs) from the PC to the server i was getting indication of 19-22MB/s , while when i tried to copy from the server back to the PC the speed was 21-24MB/s. Are these good values ?
Any ideas why the speed is different from one side to the other?
April 19, 2006 1:40:58 AM

Quote:
OK, i got the Linksys Gigabit switch and i did some testing today with the help of NetMeter Beta v0.99.

On one side i got my IBM xSeries 226 server with a Broadcom gigabit ethernet running Win2k3 Server and on the test PC a onboard Asus Nvidia NF4 Gigabit ethernet. Both machines use good Cat5e cable (Belkin)

Copying a DVD image (big file ~5gigs) from the PC to the server i was getting indication of 19-22MB/s , while when i tried to copy from the server back to the PC the speed was 21-24MB/s. Are these good values ?
Any ideas why the speed is different from one side to the other?


These figures are probably not bad. Single IDE to single IDE, I've seen around 30 MB/s consistently, and this is consistent with the drive speed -- i.e. when the same/similar drives are running in one machine, transfer rates are around 30 MB/s, no more than say 35 MB/s. (I don't have exact figures here; these are rough numbers from memory.)

So if you have slower/older/fragmented/crowded drives, the 20-odd MB/s figure could be reasonable. Same if you have other slower hardware, etc.

The open source TTCP can tell you what your network is capable of. File transfer speeds will be significantly slower as a rule.

As to server-client and vice versa, I've noticed a consistent performance gain by pushing instead of pulling. I.e. copying from machine A to machine B gives better results if the copy is initiated from machine A instead of machine B. I don't know why. I could speculate, but that's probably worthless. Lots of smart people around, maybe someone would actually know.

One factor that does apply however, is the cache. When files are transferred to the destination, they are placed in the in-memory file cache. So when this file is transferred back, it is conceivable that some of that file is being read from cache. This factor is diminished but not eliminated as files get very large. Even with 4.5 GB files with a cache of 1-2 GB, I've seen transfer speed improvements at the onset. You'd think that the file cache would be complete rolled over with a file this big, but that didn't seem to be the case.

Another factor could be the drive speeds themselves. One drive might be slower than the other. Writes should typically be slower than reads (notwithstanding some "cheating" due to write caching and possible variations in the "I'm done!" time). Bah. Too much speculation. Figure out what's (consistently) faster and don't worry about it..
May 30, 2006 1:08:56 AM

Best Gigabit switch (for the money):

Linksys (relabelled CISCO) 24-port 10/100/1000 layer3 manageable with 802.3ad support. Linksys SRW2024.

http://www.linksys.com/servlet/Satellite?c=L_Product_C2...

I've found it for about $500 CAD. Its very, very sexy for that price. CISCO awesomeness without the cisco price.

It supports 802.3ad trunking (port aggregation) = the win!
June 5, 2006 7:01:39 PM

Anyone has experiences/opinions/facts about the Dell PowerConnect 5324 (24 ports gigabit) ?
http://www.dell.com/content/products/productdetails.asp...

and it's relatively pretty cheap as well.
I already own 3 at work. looking to get 2 more gigabit switches to complete my network. Since this segment is a little bit more sensitive and important, i am trying to find out if i could use those PowerConnect still.

Dell 5324 = ~1200$ canadian. ( 24 ports )
Cisco cataclyst 2960 = ~5400 canadian ( 48 ports )

looking for options/possibilities.
June 6, 2006 5:50:05 PM

If this is for a business application Cisco would be my first choice. My second choice would be Extreme Networks.
June 6, 2006 9:13:02 PM

I think I stated this in some other thread......I have a 24 port netgear gig switch. Its web-managed, 2 SFP slots, suports VLAN's etc......$330 I thought it was a pretty good deal. The comparable Linksys one didnt even support VLAN's so I bought this one instead.....
!