My test bench

[System Error Message]

Here i will introduce my current test bench. I will update this thread as things change and include pictures later.

This is how the test will be done as close as possible. Its very much like @thiggins setup but with more test servers and a seperate network for monitoring and control of servers and test device.

[servers]
[1] AMD phenom ii x3 unlocked to x4 and overclocked to 3.2Ghz, 16GB ram,mellanox SFP+ card ,opensuse (software not ready)
[2] Intel x5650 overclocked to 4Ghz, 16GB ECC ram (cant find the other stick),dual SFP+ card, opensuse (software not ready, NIC issue)
[3] AMD piledriver 8 core 4.3Ghz, 32GB ram, Intel server quad port NIC, windows server 2012 (software not ready)
[4] AMD bulldozer 4 core, 8GB ram, Intel server quad port NIC (I350) sun branded, (no OS)
[5] another board for intel x5650 (CPU socket pins bent)
[6] 2x L5520 (cant find on the hp server board where the pins for power button is, really its that difficult)
[7] quad core i7 ivybridge laptop, 12GB ram, aetheros NIC
[8] core2duo, (still in assembly -PSU, HDD)
[9] core2duo thinkpad, (needs assembly from bits.)
[10] core2duo laptop, (no OS)
[e] Prodigit Electronics 2000MU wattmeter

Ofcourse the servers do other tasks while not benchmarking, such as being a server for web, games, media, files and warming up my room. They do have to pull their weight and earn their upkeep.
[software]
Software that is used for testing.
-netdata
-iperf
-apache server (apache bench)
Currently i dont have a clue what i should use, so please write your suggestions
[Testing]
Stuff that is tested.
- ipv4 NAT performance (both hardware acceleration and no hardware acceleration) (64B, 512B,1500B)
- ipv6 static routing performance (both hardware acceleration and no hardware acceleration)
Do suggest more. (64B, 512B,1500B)
- ipv4 NAT with PPPOE (64B, 512B, 1500B)
- apache server file benchmark over ipv4 NAT (100,1K,10K,100K of 10KB, 1MB file)

So out of all the routers i have on hand right now, try to guess which brand will go first
Mikrotik
Ubiquiti
Asus
Zyxel
BT homehub 5A.

Feel free to discuss/criticise
For future tests QoS can be considered but the software to use and router configuration files will need to be made for quicker test setups.

 

[sfx2000]

So what about the test cases themselves?

Lot of gear on the bench - what are you trying to sort out?

Performance, capacity, throughput?

 

[System Error Message]

So what about the test cases themselves?

Lot of gear on the bench - what are you trying to sort out?

Performance, capacity, throughput?

Nope, just trying to sort out some of the hardware to get them running and sort out the software to run on them. In terms of hardware thats ready i have the performance and capacity to test the CCR1036 already but need to install software and configure.

The test case was one of the things i was asking, what test cases would you like to see?

 

[sfx2000]

Remind me tomorrow - and building up test bench here is nice...

 

[System Error Message]

I've added the test used by ars technica if it is using the apache web server benchmark for downloading files. I need suggestions on what to test.

 

[kvic]

I've added the test used by ars technica if it is using the apache web server benchmark for downloading files. I need suggestions on what to test.

Personally I think the Ars test is a very good starting point. I would go straight and test it on Asus...can't wait to see your first results!

 

[sfx2000]

Personally I think the Ars test is a very good starting point. I would go straight and test it on Asus...can't wait to see your first results!

Actually, I do not think those test cases on ARS are very good... as it's just a single client, and it's pretty easy to fit test cases to a specific implementation...

One needs to have a fair number of clients - let's say 20, and then characterize different traffic flows - VOIP, Video, Best Effort, Bulk Transport, and then let's see what works and what doesn't...

Considering where we are at the moment - QoS challenges are many, and total capacity is the true limit of a router - not just how many sessions from a single client.

That'll separate the real routers from the pretenders...

 

[kvic]

Actually, I do not think those test cases on ARS are very good... as it's just a single client, and it's pretty easy to fit test cases to a specific implementation...

One needs to have a fair number of clients - let's say 20, and then characterize different traffic flows - VOIP, Video, Best Effort, Bulk Transport, and then let's see what works and what doesn't...

Considering where we are at the moment - QoS challenges are many, and total capacity is the true limit of a router - not just how many sessions from a single client.

That'll separate the real routers from the pretenders...

True. That probably also separate real developers from the pretenders...

I'll hold off a little bit to be pragmatic and see how SEM goes with the Ars test and avoid drilling down the path of a test spec drafting effort before the project even boots up.

 

[System Error Message]

well so far i've had a few issues setting things up. Configuring linux in a headless manner is very very difficult, aside from that i also need to work on "how to manage a home made cluster" involving central configuration methods too, as i said the computers need to earn their upkeep so thats one of the things being worked on at the same time. Just after doing a GPU swap the piledriver system wont boot up as windows server hated the GPUs i had in before.

With QoS it would be nice to test but the setup would be tedious as it would involve router config and setting up the applications to test it as i havent yet gotten remote desktop working and teamviewer doesnt seem to work as it keeps hanging.

Trying to set up a GPGPU cluster or even a standard cluster is difficult if you build it out of what you can cheaply find but i will have it mostly online soon enough.

So can anyone guess which brand will go first?

 

[System Error Message]

Not far from being ready. All thats left is to install opensuse on the bulldozer 4 core and install the software on all of them. I'll try finding another mellanox 10G NIC cheap to replace the Qlogic ones that refuse to work.

 

[sfx2000]

Configuring linux in a headless manner is very very difficult,

It's actually _not_ difficult - just gotta know how it's done - com1 is your friend

With the HP server main board - connect ethernet to the service processor port, and that'll give your ILO - there's a default IP for ILO that one can http into which will give you KVM capabilities (it's java based, and old Java at that...)

 

[sfx2000]

Just after doing a GPU swap the piledriver system wont boot up as windows server hated the GPUs i had in before.

the window server might die, but it is actually most likely booting, so you can either SSH in, or use the console...

PM me and we can discuss there...

 

[System Error Message]

the window server might die, but it is actually most likely booting, so you can either SSH in, or use the console...

PM me and we can discuss there...

Not to worry, i sorted out and overclocked the windows server without an issue. Also configured LACP on it as well using the intel NICs and the nvidia GPU drivers installed without any issue. GTX 580s arent as power efficient as the HD5800s as i plan to leave it running 24/7 but my choice for GPUs was for compute support and the older gens arent as aggressive in restricting double FP performance as the new gens are.

The problem was caused by the CPU being a bit unstable with a 100Mhz overclock and also with a usb drive plugged into the usb3 during boot. Seems to not like booting with usb3 and EHCI handling enabled in BIOS so i did a BIOS reset. However since i was going to overclock regardless i raised the voltage up by 0.1V to 1.4V and went straight for 4.3Ghz with 4.5Ghz turbo and stress tested it to be stable. Piledriver max safe voltage is 1.55V so the one i had which wasnt the famous fx8350 (its just the fx8320) so it seems to have been configured for a lower TDP. I may be able to get another 100Mhz or 200Mhz out of the CPU on the same volts. Every extra CPU performance i can squeeze will help with the benchmark and in running day to day tasks.

One of the GTX 580 i have doesnt seem to show any display from bios so i swapped the positions of the 2 GTX 580s.

Im dealing with hardware i can scavange cheap. Some of the hardware i have are of excellent quality and can overclock really far or perform very well while some will only work in certain conditions. I only have 1 server board, rest are all desktops so COM1 is not useful.

If the windows server did boot before my router would pick it up.

 

[sfx2000]

I only have 1 server board, rest are all desktops so COM1 is not useful.

If the windows server did boot before my router would pick it up.

tty0 - which is typically the console - can be mapped out to COM1, or to a USB port - normally USB0, and then one can use a USB/Serial connection to get to the console (every linux hacker should have a USB/Serial-RS232 adapter in their toolkit) - and if the window server fails to launch, CRTL-ALT-T will get you there with Ubuntu and Mint, with RedHat, I'd have to double check, but if I recall, it's CTRL-ALT-F1 that will get you into the console, and just login as either root, or a user with sudo...

I'm not sure about how Windows Server does this, but if I recall, it's pretty dependent on ILO there (virtual KVM), at least what I've gleaned from our Windows Server wrangler...

 

[System Error Message]

tty0 - which is typically the console - can be mapped out to COM1, or to a USB port - normally USB0, and then one can use a USB/Serial connection to get to the console (every linux hacker should have a USB/Serial-RS232 adapter in their toolkit) - and if the window server fails to launch, CRTL-ALT-T will get you there with Ubuntu and Mint, with RedHat, I'd have to double check, but if I recall, it's CTRL-ALT-F1 that will get you into the console, and just login as either root, or a user with sudo...

I'm not sure about how Windows Server does this, but if I recall, it's pretty dependent on ILO there (virtual KVM), at least what I've gleaned from our Windows Server wrangler...

Wasnt the issue, i had simply forgotten my password and did the hackish password change after i had reset the bios.

 

[sfx2000]

Wasnt the issue, i had simply forgotten my password and did the hackish password change after i had reset the bios.

Cool - anyways, you know more perhaps than what you're telling us - which is ok, as you're trying to bring things up...

Just have a plan, and know that Linux at least, does have options for headless operation...

 

[kvic]

Looks promising..

Can't wait to see apache benchmark up and running. There might be a little challenge to graph the bandwidth charts.

 

[System Error Message]

back to the fun bit, Can anyone guess which brand will go first?

 

[L&LD]

back to the fun bit, Can anyone guess which brand will go first?

If I've gotten to know you at all during the last few years, I'll guess 'Mikrotik'.