jeremymacmull

Gigabit Switches

Recommended Posts

Damn the American Dell deals  :( .  You guys don't know how lucky you are.

Sometimes I wonder if many Canadian geeks consider moving to the United States just for the Dell deals.

206139[/snapback]

I'm one of the Canadian people, I get things shipped to one of the US facilities owned by my employer. Biggest problem is convincing places to ship using a canadian credit card.

Share this post


Link to post
Share on other sites
Damn the American Dell deals  :( .  You guys don't know how lucky you are.

Sometimes I wonder if many Canadian geeks consider moving to the United States just for the Dell deals.

206139[/snapback]

I ordered 2724 too. I was getting 50MB/s with 2616 already. Now both my server and workstation are using RAID which can read/write at over 120MB/s. With this jumbo frame supported switch, I want to see how fast I can trasfer data.

BTW, it was on sale in Canada too so I ordered it.

Share this post


Link to post
Share on other sites
Damn the American Dell deals  :( .  You guys don't know how lucky you are.

Sometimes I wonder if many Canadian geeks consider moving to the United States just for the Dell deals.

206139[/snapback]

Quite, Dell as usual are ripping off UK consumers, Here the 2708 is $143 + 17.5% VAT, the 2716 is $289 + VAT.

Bleeding typical.

greg

Share this post


Link to post
Share on other sites
I ended up ordering the 2724.  It does have one fan.  I can't believe I got a 24-port managed 10/100/1000 switch with vlan, jumbo frame, tos, link aggregation, etc., etc. for under $300 after tax and shipping.  I'll report back when I get it, which should be 1st week of June.

206138[/snapback]

Dell are charging $544 + 17.5% VAT for that here in the UK.

greg

Share this post


Link to post
Share on other sites
cmkrnl wrote:

The 16 & 24 port versions of the new Dell 27xx switch range have both vlan and Jumbo frame support.

http://www1.us.dell.com/content/products/c...s=04&l=en&s=bsd

http://www1.us.dell.com/content/products/p...s=04&l=en&s=bsd

Dunno WTF they disabled jumbo support in the 8 port one. I assume they are based on all the same switching hardware.

Greg

I've been looking around, and now I see why our overseas friends are 'complaining' about the Dell deals. :) The closest I could come to on the 16 port model is this from Netgear: http://www.netgear.com/products/details/GS716T.php

That has Jumbo Frame support up to 9,600 byte packets. But, that firmware isn't available now and the switch itself is 2X the cost. Netgear's 16-port model does have the fiber option though, where Dell has it in the 24-port version only.

Does anybody see a need for jumbo frames greater than 9,000 bytes? If not, I think I'm going to go for the 16-port Dell model. There is no fan to break, and I am not likely to need a fiber connection in my house or more than 16 connections. :)

Share this post


Link to post
Share on other sites
BTW, it was on sale in Canada too so I ordered it.

206155[/snapback]

Ya, I immediately checked the Dell.ca price since it looked like a nice switch at a reasonable price. It was 15% off if my memory served me right, but was still going to be over CAN$300 before shipping. At ~CAN200-250 I would have gotten it, but quite frankly it's a little overkill for my home network. I could buy several SMC8505T's for that price.

Share this post


Link to post
Share on other sites
So what's a router? Is it a switch or a hub?

206383[/snapback]

Neither. A router is a device that provides IP routing and/or NAT and thus technically nothing to do with switches or hubs.

Some manufacturers build hubs into their routers. Some build switches into their routers. Some build wireless access points. Some don't do either.

Share this post


Link to post
Share on other sites

I've been surfing around a bit, trying to figure out this Jumbo Frame thing and the VLAN concept to supposedly allow mixed frame sizes on the same switch. From what I see, with the switches that we all can afford (like Dell, Netgear, SMC, Linksys, etc.) I don't think that we will be able to have both Jumbo and non-Jumbo frames on the same switch any day soon.

The reason is that we can create a VLAN on a particular switch for jumbo vs. non-jumbo, but a switch by itself then has no way to route those packets from one port to another. All you do is isolate ports from other ports. You need a switch that has a routing feature in it, otherwise a VLAN would isolate one part of your network from the other...which is not what you want to do here.

See this Dell explanation below that I found. Does it make sense or is this old information? Based on this, it may be some time until we see Jumbo and non-Jumbo switches in one unit. This does not appear to be as simple as combining 10BT/100BT into one switch as was done years ago. In that case, we were just adjusting packet speeds and not having to combine/break apart packets. Hmmm.

http://www.dell.com/downloads/global/produ...app_note_38.pdf

As noted above, jumbo isn't everything, but they sure do make it tough. Correct me if this is wrong. :)

Share this post


Link to post
Share on other sites

Oh ya, there's one more possibility. Get a computer with two NIC cards in it, set up the VLAN, and use the computer to convert jumbo to non-jumbo between switch ports. Either way, it doesn't sound too easy...

Share this post


Link to post
Share on other sites
Oh ya, there's one more possibility.  Get a computer with two NIC cards in it, set up the VLAN, and use the computer to convert jumbo to non-jumbo between switch ports.  Either way, it doesn't sound too easy...

206406[/snapback]

You dont need two NICs to do it, you need just one to configure what's termed a 'router on a stick'.

It doesnt get any easier.

greg

Share this post


Link to post
Share on other sites
Oh ya, there's one more possibility.  Get a computer with two NIC cards in it, set up the VLAN, and use the computer to convert jumbo to non-jumbo between switch ports.  Either way, it doesn't sound too easy...

206406[/snapback]

Based on this info for home use, it probably does not pay for us to get on the "Jumbo frame" band wagon :) this early. Unless someone says my previous assumptions where wrong and has a 2716 or 2724 to verify it, I'm tempted to change my order to the 2708 tonight. It only pays to get a cheap one for now and upgrade in the future. The 8-port SMC is about the same price, so it's a trade-off between a managed non-jumbo and a non-managed "all or nothing" jumbo.

Share this post


Link to post
Share on other sites
Oh ya, there's one more possibility.  Get a computer with two NIC cards in it, set up the VLAN, and use the computer to convert jumbo to non-jumbo between switch ports.  Either way, it doesn't sound too easy...

206406[/snapback]

Based on this info for home use, it probably does not pay for us to get on the "Jumbo frame" band wagon :) this early. Unless someone says my previous assumptions where wrong and has a 2716 or 2724 to verify it, I'm tempted to change my order to the 2708 tonight. It only pays to get a cheap one for now and upgrade in the future. The 8-port SMC is about the same price, so it's a trade-off between a managed non-jumbo and a non-managed "all or nothing" jumbo.

206461[/snapback]

Speaking from my experience, on my non-jumbo-frame switch, without OS tuning, wire speed is arround 300mb/s. After tuning, it's 950mb/s. It may reach 980mb/s if the switch supports jumbo-frame, but you can see here TCP tuning is a lot more important than jumbo-frame. This is why people see 25-30MB/s network trasffer speed and are disappointed. I was very disappointed until I learned how to optimize TCP.

The reason to go with 2700 series is not for jumbo-frame, it's managed features like vlan, lag, qos, etc. Without jumbo-frame, 2708 is still a much better buy than smc.

Share this post


Link to post
Share on other sites
Could you share how you tuned it ?  :D

206505[/snapback]

Very simple

I use iperf to find out best TCP window size for my connection, also monitor CPU load, find a best combination.

Then I use doctorTCP from http://dslreports.com to setup, no need to manually edit registry.

You can see from flowing, TCP window size makes huge difference. but if you set it too high, CPU load will be very high too.

C:\Documents and Settings\Administrator.US>iperf -s -w 8k

------------------------------------------------------------

Server listening on TCP port 5001

TCP window size: 8.00 KByte

------------------------------------------------------------

[1896] local 192.168.0.100 port 5001 connected with 192.168.0.50 port 1505

[ ID] Interval Transfer Bandwidth

[1896] 0.0-10.0 sec 369 MBytes 309 Mbits/sec

C:\Documents and Settings\Administrator.US>iperf -s -w 16k

------------------------------------------------------------

Server listening on TCP port 5001

TCP window size: 16.0 KByte

------------------------------------------------------------

[1896] local 192.168.0.100 port 5001 connected with 192.168.0.50 port 1506

[ ID] Interval Transfer Bandwidth

[1896] 0.0-10.0 sec 377 MBytes 316 Mbits/sec

C:\Documents and Settings\Administrator.US>iperf -s -w 32k

------------------------------------------------------------

Server listening on TCP port 5001

TCP window size: 32.0 KByte

------------------------------------------------------------

[1896] local 192.168.0.100 port 5001 connected with 192.168.0.50 port 1507

[ ID] Interval Transfer Bandwidth

[1896] 0.0-10.0 sec 634 MBytes 532 Mbits/sec

C:\Documents and Settings\Administrator.US>iperf -s -w 64k

------------------------------------------------------------

Server listening on TCP port 5001

TCP window size: 64.0 KByte

------------------------------------------------------------

[1896] local 192.168.0.100 port 5001 connected with 192.168.0.50 port 1508

[ ID] Interval Transfer Bandwidth

[1896] 0.0-10.0 sec 1023 MBytes 858 Mbits/sec

C:\Documents and Settings\Administrator.US>iperf -s -w 128k

------------------------------------------------------------

Server listening on TCP port 5001

TCP window size: 128 KByte

------------------------------------------------------------

[1896] local 192.168.0.100 port 5001 connected with 192.168.0.50 port 1509

[ ID] Interval Transfer Bandwidth

[1896] 0.0-10.0 sec 1.08 GBytes 930 Mbits/sec

C:\Documents and Settings\Administrator.US>iperf -s -w 256k

------------------------------------------------------------

Server listening on TCP port 5001

TCP window size: 256 KByte

------------------------------------------------------------

[1896] local 192.168.0.100 port 5001 connected with 192.168.0.50 port 1510

[ ID] Interval Transfer Bandwidth

[1896] 0.0-10.0 sec 1.09 GBytes 934 Mbits/sec

Share this post


Link to post
Share on other sites

Cool real-world testing dilidolo. Well, if nothing else we've put a lot of confusing information about Gigabit and Jumbo frames to rest. If I find out any other information I'll be back, otherwise happy switch buying everyone. :)

Edited by Krusher

Share this post


Link to post
Share on other sites
Cool real-world testing...

206513[/snapback]

IPerf isn't "real-world testing" by any means. It is a useful diagnostic tool, but it's not real world.

A real world test of a GbE network would necessarily involve a more complicated setup with more potential bottlenecks (unless the whole point of your network is running IPerf :blink: . A real real-world test would be SMB or NFS file transfers. That would actually be something that a network is used for in the "real world."

Does it bother anyone else that "real-world" gets tossed around so much these days. You hear in marketing that "3DMark05 is real world." WTF!? Who builds a computer to run 3DMark05?

Share this post


Link to post
Share on other sites
days.  You hear in marketing that "3DMark05 is real world." WTF!?  Who builds a computer to run 3DMark05?

206550[/snapback]

Perusing some of the more 31337 'overclockers' forums, one could draw that inference at some of the silliness engaged within.

greg

Share this post


Link to post
Share on other sites
Cool real-world testing...

206513[/snapback]

IPerf isn't "real-world testing" by any means. It is a useful diagnostic tool, but it's not real world.

A real world test of a GbE network would necessarily involve a more complicated setup with more potential bottlenecks (unless the whole point of your network is running IPerf :blink: . A real real-world test would be SMB or NFS file transfers. That would actually be something that a network is used for in the "real world."

Does it bother anyone else that "real-world" gets tossed around so much these days. You hear in marketing that "3DMark05 is real world." WTF!? Who builds a computer to run 3DMark05?

206550[/snapback]

May I ask where in my post I said I was benchmarking my real world network performance? Iperf tests wire speed, not real work performance.

Iperf helps you tune your TCP stack. If your wire speed can not reach 500mb/s, how can you reach 50MB/s CIFS/NFS speed? I post the result only to show how TCP window size affects wire speed and this is the reason you need to tune it.

I'm curious how you find you netowrk bottleneck and what tools you use to tune your gigabit network. Dear to share?

Share this post


Link to post
Share on other sites

dilidolo, I was addressing Krusher's comment, not your post ;).

If you read my post, you will see that I note that Iperf is indeed a useful diagnostic tool, just as you also note...

Share this post


Link to post
Share on other sites

And I was more or less implying that dilidolo did a "real-world" test of the switch performance, not the application layer performance. His tests showed that you can get near gigabit speed with a non-jumbo switch. Unless I am missing something here.

Share this post


Link to post
Share on other sites
And I was more or less implying that dilidolo did a "real-world" test of the switch performance, not the application layer performance.  His tests showed that you can get near gigabit speed with a non-jumbo switch.  Unless I am missing something here.

206614[/snapback]

The bottleneck wasn't switch performance as you noted, but host performance.

Share this post


Link to post
Share on other sites
And I was more or less implying that dilidolo did a "real-world" test of the switch performance, not the application layer performance.  His tests showed that you can get near gigabit speed with a non-jumbo switch.  Unless I am missing something here.

206614[/snapback]

The bottleneck wasn't switch performance as you noted, but host performance.

206621[/snapback]

Sorry, guys. It's fun to be here and it's fun to share knowledge. :rolleyes:

My LSI Megaraid 320-4X and Dell 2724 are coming tomorrow, I'll let you know what real world performance I can achieve with these beasts. The file server is running RAID-10 on 4 IDE disks (OS is on Raptor) and the client will be running RAID-10 on U320 SCSI disks.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now