Gigabit Reviews & Benchmarks

[MikeDub83]

Take a look at the Intel Gigabit cards. Can anyone tell the difference between the "Pro/1000 MT Desktop Adapter" and the "Pro/1000 T Desktop Adapter"? Looks like different chipsets, but which one is better?

 

[gunrunnerjohn]

Don't know which are better, the gigabit cards I'm using are the Intel Pro/1000 MT Desktop Adapter, version C38676-001. UPC# 735858162203

 

[onelin]

Yes, gigabit does come integrated in several motherboards. One of the key features of 865/875 motherboards from Intel is that there is a dedicated bus for the gigabit...it isn't stuck on the PCI bus. I have one of these boards (the first micro ATX that came out w/ it), and I have achieved 25MB/sec sending and receiving with other users who had PCI cards (good ones) over dell and other nice switches.

It's definitely much faster than 100Mbit, which usually you get around 7-9MB/sec with a really good setup getting the full 11-12MB/sec... however, it's also hardware limited (hard drive speeds) IMO. You won't see 10x increase but with good hardware 2x-3x and maybe a bit more would not surprise me at all with average new hardware. Oh, make sure your cables don't becoming the limiting factor, too. If they couldn't max out 100Mbit they sure as hell won't do well with gigabit.

I haven't upgraded my home network to gigabit equipment yet, so it's mainly at LAN parties.

 

[jonny13]

I just upgraded my network to gigabit about 3 months ago and have been impressed with the results. I took a few screenshots of a test transfer I just did. The switch is an 8 port Linksys switch. The client machine is a 3.0C P4 with onboard gigabit via CSA (bypassing the PCI bus for faster transfers) and 512 MB RAM. The server has a 1.6 GHz P4 with an Intel gigabit PCI card and 1024 MB RAM. The biggest thing I can see is that RAM helps alot. When I transfer from my computer (Client) to the server, my memory drops to nothing, so an upgrade to 1 GB would help out alot. Anyway, here are the results.

Client to Server - averages around 20-22 MB/sec

Server to Client - averages around 47-49 MB/sec

Obviously this isn't an exact test, but these results mirror what I have been seeing over the past few months.

Jonny

Edit: Added link to the switch.
Linksys Gigabit switch

 

[JackMDS]

Wow!

Thanks jonny13

We really need this kind of Posting:beer:

Outcome like this justify upgrade to Giga.

Do you have, by any chance, a similar test with another client computer using regular PCI Giga?

I wonder what is the contribution of the Client Giga Bypassing the PCI Bus?

BTW. What is the Operating Systems Used?

 

[jonny13]

No, I actually don't have another PCI card, the only one used is in my server which I don't ever shut down. The other comp in my apartment is another one with onboard Gigabit, although that one doesn't have CSA and just rides on the PCI bus. So, I could check that one out. It only has 256 MB of RAM though, but it might be interesting to see. I will try to go a more thourogh benchmark and put it up tomorrow morning. Also, the client used Windows XP and the server was running Windows Server 2003. Both hard drives used had 8 MB buffers, but I do have a Rapter in the client I can use to see if that changes things.

Jonny

 

[JackMDS]

Big Thanks.

In the next few years the combo Win2003 and WinXP will be the Dominant Combo in small settings.

Good to know that such level of Transfer is already possible.

 

[MikeDub83]

jonny13, thanks for the benchmarks! Basicly what you're showing is that the file transfer is pushing the limits of hard disk speed. In reality, this is all you can ask for, a network transfer as fast as your hard disk can write it. Unless of course you have a sick RAID-0 array setup and you can knock on the 125 MB/s door, the theoretical bandwidth of Gigabit.

 

[gunrunnerjohn]

I have always suspected that a true server O/S would be the silver bullet for gigabit transfers, I'm considering a test system to prove or disprove that in my case. I know that with 2K and XP workstations, gigabit is about half his numbers for each type of transfer. Since writing backups from the local PC is my goal, the minimal gain is disappointing to me.

 

[jonny13]

gunrunnerjohn: I am not so sure it is just the OS that caused that big of a difference. I am leaning towards the amount of RAM the server compared to the client. When I transfer from my PC to the server, my amount of RAM left is in the double digits, and low double digits at that; even dropping to the single digits for seconds at a time. But, going from the server to the client, the server always has enough RAM available to continue making the transfer. I wish I had some more RAM laying around to add to my desktop to test my theory, but I don't at this time. I do think the server OS would be better, I just don't think it is THAT much better. Just my .02 though.

Jonny

 

[spidey07]

Even the big honkin Sun boxes we have can barely push a full 1000 Mbs. And we're talking 8 processor, 32 gig ram, etc boxes.

 

[onelin]

Originally posted by: spidey07
Even the big honkin Sun boxes we have can barely push a full 1000 Mbs. And we're talking 8 processor, 32 gig ram, etc boxes.

but regardless, there's plenty of gain on desktop systems over gigabit when compared to 100Mbit. 2-4x, even if 10x is unrealistic.

 

[gunrunnerjohn]

Both of my workstations have a gigabyte of RAM, so I really don't see this as an issue transferring a 700mb file. The speed clearly seems limited by something other than RAM. Also, I run QCheck and it indicates 500mbit or better throughput both ways, so I seem to have enough bandwidth available to do better than I'm seeing. The curious part is the difference between reading and writing disk files from the machine, that indicates it's something about remote writing that is sucking up a lot of time.

 

[Kilrsat]

Some numbers from today:
Server:
P3-1000
1GB pc133
Intel Pro/1000 MT server adapter in a 64bit 66mhz pci slot.
4x36GB scsi drives in Raid-5
Windows Server 2003

Client 1:
P4-2.8Ghz
2GB pc2100
Integrated Intel gigabit adapter
80GB ide harddrive
Windows XP Pro

Client 2:
Dual Xeon 2.8Ghz
2GB pc3200
Integrated Intel gigabit adapter
3x36GB scsi drives in Raid-0
Windows XP Pro

Client 3:
Dual Xeon 1.4Ghz
2GB pc133
Onboard Broadcom gigabit adapter
4x72GB scsi drives in Raid-5
Windows Server 2000

Basic test:
Copy a folder containing 4.12GB of information (128 files, 15 subfolders) from the server to each client.

Peak transfer:
76.1MB/s

Sustained transfer:
55MB/s


Granted these are fairly high-end machines, but my testing is for the purposes of pushing our gigabit network to the limit. And I'm not even sure I've done that yet. I'll be doing some write testing, simultaneous read+write, adding in some low end clients, and some between client transfer tests later this afternoon.

If there is some specific test any of you guys would like to see, let me know. I should have the hardware available to do just about anything.

*EDIT*
Early write results:
I had the 3 clients write the 4GB folder back to the server (seperate destinations of course) at the same time. With a single client it peaked at 40.2MB/s, but held a steady 25MB/s. With multiple clients it peaked at 44.6MB/s, with a steady 35MB/s. I think we're maxing out the write performance on this older raid-5 array. I'll get to testing a newer raid-5 array later, along with multiple clients to a 3 x 15k drive raid-0 array.

 

[JackMDS]

Kilrsat, thanks Nice helpful Post.

I would appriciate if you can measure the ?Speed? that the Client one (with Win XP)writes to the Server to as compares to the other clients.

You have nice setting and client one is the closet to what most of us are using.

John

Getting this results form Qcheck takes out of the ?BAD? loop the Cables Switch and small part of the OS contribution.

That means that the slowness can be a combo of the parts of the OS the Mobo and Hard drive storage system.

As a next step you can try to take out of the loop the hard drive by using 200-300 MB of you RAM as a Virtual disk and save the transfer to the virtual Disk.

Unfortunately I can not contribute to the testing since my Giga system is Iced out (I.e. it in a location that need a Ferry boat to get to, and Ferry isn?t running since the water froze). :brokenheart:

 

[Kilrsat]

Used another system.
Client 4:
Dual Xeon 2.8Ghz
1GB pc3200
Integrated Intel gigabit adapter
80GB IDE harddrive
Windows XP Pro

From Client 4, pushing to Client 1:
Peak: 41.9MB/s
Average: 25MB/s

Looked like it was sustaining 35MB/s fairly easily during the large file section of the copy, but it was the many folders with many 10MB files that brought the overall rate down.

From Client 1, pushing to Client 2:
Peak: 46.8MB/s
Average: 30MB/s

Was pretty consistent through both the large files and the folders.

I'm thinking about dropping a gig card into a P3 500. Anyone want to see those numbers?

*EDIT - More "pushing" numbers*
Multiple clients pushing to Client 2 (using integrated adapter):
Peak: 46.1MB/s
Average: 30MB/s

Multiple clients pushing to Client 2 (using Intel Pro/1000 MT Server Adapter in 64bit PCI slot):
Peak: 52.1MB/s
Average: 31MB/s

Looks like the integrated one is up to snuff. Going to get the P3-500 test ready.

 

[gunrunnerjohn]

Yep, numbers on a vastly slower machine could point to network and bus overhead and not disk overhead.

BTW, are you using jumbo packets on your gigabit links? If you are, please disable them for a test, I'd like to know how much effect that would have.

 

[JonnyBlaze]

the fastest i have ever sent on my 100mbit lan is 10.5MB/s

so the number about 2~4 times faster seems right. i bet theres more improovment when the network has even more pc's on it.

JB

 

[Kilrsat]

Originally posted by: gunrunnerjohn
Yep, numbers on a vastly slower machine could point to network and bus overhead and not disk overhead.

BTW, are you using jumbo packets on your gigabit links? If you are, please disable them for a test, I'd like to know how much effect that would have.

All tests so far have been with jumbo packets disabled. Setting up a baseline mark before I really start changing anything. No OS changes, just drop in the network card, give it the latest drivers, and let the defaults chug away.

*EDIT - P3 500 results*
Client 5:
P3-500
384mb pc100
Intel Pro/1000 MT Server Adapter (in 32bit pci slot)
10GB ide harddrive
Windows 2000

Pulling from Server:
Peak - 13.5MB/s
Average - 11MB/s

Pushing to Server:
Peak - 18.6MB/s
Average - 14MB/s

More tests will have to wait until I go back to work tomorrow. However, if you have something you would like to see just leave a note and I'll try to get it done then.

 

[MikeDub83]

All tests so far have been with jumbo packets disabled. Setting up a baseline mark before I really start changing anything. No OS changes, just drop in the network card, give it the latest drivers, and let the defaults chug away.

I'd like to see a benchmark with jumbo frames enabled.

 

[gunrunnerjohn]

It's interesting that the P500 test has the push of files faster than the reading from the server. In all my tests, it works the other way. I'm planning on installing Server 2003 on a system and seeing if it's really as simple as having a server O/S or maybe something more basic...

 

[Kilrsat]

Originally posted by: MikeDub83

All tests so far have been with jumbo packets disabled. Setting up a baseline mark before I really start changing anything. No OS changes, just drop in the network card, give it the latest drivers, and let the defaults chug away.

I'd like to see a benchmark with jumbo frames enabled.

Busy day, but I spent the morning looking into this and apparently our HP ProCurve switch (4108gl, if you care) doesn't support jumbo frames. (There seems to be only one HP model that does, damn). So unless I can get another switch onhand, I won't be able to do the jumbo frames tests.

 

[gunrunnerjohn]

Bummer, I was thinking of replacing my switch (which also doesn't support jumbo frames) with the SMC 8 port model that does, but I'd like to know if I'm wasting my money.

 

[cmetz]

Kilrsat, you can always cheat and use a crossover cable. Jumbo frame support is still spotty, unfortunately.

 

[foshizzle]

I'm thinking of hooking 2 comps together with a X-over cable with 2 cheapo gigabit cards. They're only like 20 bucks at Newegg. Do you guys think that they will have decent performance with a X-over?