Network Benchmark / Testing

[dastrix]

Hi All,

 

We have aging desktops and laptops, however one of my guys keeps insisting that the slowness is a problem with a switch somewhere, so i am going to run some benchmarks to try and quash the so called problem.

 

We have a HP 3800 Core switch and Procurve 2910 and 2910 poe switches. Everything is gigabit at the client.

 

Can anyone recommend some decent ways to test speeds and what sort of speeds we should get from this kit generally?

 

Cheers

[Steve21]

What is the "slowness" referring to? If it's logons for example could be your switch uplinks don't cope with 30 machines suddenly pulling down appdata etc etc.

 

If it's web-browsing could be internet etc etc

 

Bit hard to know what to suggest without knowing more And speed in relation to....

 

Steve

[dastrix]

its not logins and its like getting blood from a stone from my guy when asking so i just wanted to benchmark and then we can compare different blocks etc.

 

We have got slow internet, and tied in for ever with BT so sitting that one out.

[Steve21]

Benchmark what though :s I mean like if you're talking how long it takes to load Photoshop etc, it'll be different to how long it takes to download XYZ. Or whether it's how long a certain file takes to open etc.

 

Steve

[Sagima]

I suppose what you might be looking for is a switch that has autonegotiated to 100meg rather than a gig (or half duplex instead of full) and as it's not a specific client that is slow could you not just log onto each switch and set the uplinks to be 1gig. If that doesn't solve the problem check the file transfer speeds to each end of your network and if one is significantly slower transfer back one switch at a time until the speed is acceptable again and look at the cabling between those switches? Your client smight not read/write at 1gig but they should easily do more than 100meg

[dastrix]

Im thinking to try and get some sort of benchmark to basically move a large file over the network and then try and capture times etc

[mikeyd101]

if you want to test link speed between 2 points on a network i'd suggest iperf, this will give you max throughput of the network interface / switches.

 

I'd also be looking at port traffic information on the switches themselves, rx/tx errors, collisions and data rates etc. Do you have any uplink monitoring in place? I use "the dude" to monitor our network interlinks, mainly because its free.

[dastrix]

We dont do any uplink monitoring unless the switch does it, so using the dude do you just install that on a central server ?

 

I have just found a thing called Lan Speed Test and written and read a 2gb file and it wrote at 31,095,912 Bits Per Second and Read at 91,779,464 Bits Per Second

 

Does that sound about right?

[mikeyd101]

We dont do any uplink monitoring unless the switch does it, so using the dude do you just install that on a central server ?

 

I have just found a thing called Lan Speed Test and written and read a 2gb file and it wrote at 31,095,912 Bits Per Second and Read at 91,779,464 Bits Per Second

 

Does that sound about right?

 

I just installed it on an old laptop (power options set to never sleep), I also have it displaying an overview map sitting on my top shelf so i can see all critical hardware if it goes offline (pings), it doesn't have that high of requirements. I'd suggest playing around with on a test machine before committing it to server duty.

 

Obviously reading /writing files will give combined figures of network and hard disk throughput. That's why i use iperf to only test network transfers. But your numbers seem ok, writes are a little slow but may be restricted to disk speed that end?

 

I'm sure your switches will have SNMP abilities, our switches do fairly old extreme's and hp 56xx switches. SNMP is the way the dude extracts the transfer information/ stats from the switch this may need to be configured if its not on by default. You'll also need to know the ip addresses for all your switches, and it helps if you know what the uplink ports on each switch are. I had none of these for our network so was also a discovery mission to find out what was connected to what and how fast, and how much bandwidth was in use.

 

As I've found out network management is an alice in wonderland rabbit hole

[dastrix]

First job it to get the dude installed tomorrow then that way it can start collecting info, luckily I have all my network switches and locations etc documented

[mikeyd101]

I'd also be looking at network latency. just because you can transfer at that point in time at that speed doesn't get you the full picture.

 

The dude will also do that with pings to the hardware at x intervals. Be careful about your check intervals, more checking is more overhead (so can make problems worse) I have mine set to 30 secs so polls everything once every 30 secs. I have about 200 devices monitored.

[mikeyd101]

First job it to get the dude installed tomorrow then that way it can start collecting info, luckily I have all my network switches and locations etc documented

 

My advice, avoid discover, just create a new network map and add devices 1 by 1, start with the core and work your way out, or just concentrate on the area you want to explore to start.

use sub maps if you have alot of the same type of devices to monitor, I use that for things like our ip cameras at each campus

 

Having that historic data is a really big help and identified a few of our links that are heavily saturated (due to our ip cameras), not that the boss has done anything about it, we have the information. It also helped us force an issue with AT&T about one of our campus links to get it repaired. AT&T are BT x 100

 

Any questions give me a shout, and i'll try and help out, the best i can.

Edited January 5, 2016 by mikeyd101

[psydii]

We had some really odd performance problems for a while. Turned out a fibre patch lead was loose.

 

Look for errors on the interconnected ports of your switches.

[Davit2005]

Hi All,

 

We have aging desktops and laptops, however one of my guys keeps insisting that the slowness is a problem with a switch somewhere, so i am going to run some benchmarks to try and quash the so called problem.

 

We have a HP 3800 Core switch and Procurve 2910 and 2910 poe switches. Everything is gigabit at the client.

 

Can anyone recommend some decent ways to test speeds and what sort of speeds we should get from this kit generally?

 

Cheers

 

Are the 2910's 48 port fully populated??

 

We did have issues with our 2910's but that may of been due to the way they were uplinked, it was an inherited bit of a mess (putting it nicely) There were 2 almost fully populated 5212's. Once we ripped out the 2910's and squeezed everything into the 5212's log in speeds improved by double.

[fiberoptics]

get another switch to see if the problem still exists.