speedtest.net giving different results VM vs. PA-220

[stlenger]

after switching from a PA VM on an ESX to a PA-220 i noticed a huge decrease in throughput from speedtest.net - with the VM in vwire or l2 mode i am getting rates consistent with native line speed (around 500Mbit/s down and 250Mbit/s upload). With the PA-220 in either vwire or l2 i am seeing a huge decrease in results in the region of 60 Mbit/s down and 50 Mbit/s upload.

Both the VM and PA-220 are on 8.1 (both demo/lab machines), it doesn't matter if av/threat protection is disabled or enabled.

Curious is that other speedtests will show better results even with the PA-220 in place (nperf.net shows same speed regardless of vm/220), also real world applications will utilize full upload bandwith as far as i could test. 

My concern is that the PA-220 does handle certain traffic differently or a bottleneck is produced that slows down speedtest.net results considerable.  

Any input on this would be much appreciated!

[reaper]

speedtest uses a single session to pump as much bandwidth as possible

The strenght of the hardware platform is that it uses several different cores to process all traffic, allowing it to scan everything in-line without a cost to overall performance.

A side effect is that a single flow may have slower than full system speed performance since it will only be processed on a single core.

Nperf starts multiple sessions so the platform can distribute those evenly and get to maximum capacity.

[stlenger]

thank you for the insight! 

but two things; 

- VM has only 1 core

- PA-220 only has one dataplane cpu (or?)

so can it really be the reason? also single downloads from other sources use full bandwith. its only speedtest.net that does not...

[reaper]

the PA-220 has 2 DP cores

the vm may have a faster core, or use multithreading (or has been set up with a single multicore processor?)

you can track the cpu usage when you run the test to see how the load affects your cores: show running resource-monitor second (run a few times throughout the test)

It may also be the way speedtest sends packets. you could set up a packet-diag filter and track your global counters to see if there's a lot of fragmentation or other 'odd' behavior seen

[stlenger]

thanks for the input, i will test all that. I allready had a peak into cpu usage and couldn't see any high peaks during those tests.

The curious thing is that:

a) speedtests with nperf hit 500Mbit/s while speedtest.net hits around 60Mbit/s - both on the PA-220, single Downloads able to utilize full bandwith. 

if i put in the factor of having 2 cores and with nperf it will utilitze both cores (on the pa-220) than i should have at least 200-250Mbit/s

your suggestion with fragmentation or 'odd' behaviour is what i am counting on because everything else utilizes full speed. Its just bugging me that a VM in the same scenario and setup does not have that problem -at all- (i cant see any high utilization on VM either)