50M Speed test sites and why you shouldnt take them for granted - Page 6

qasdfdsaq · 23-02-2012, 20:07

As well as a million other things we can't exhaustively cover here...

General Maximus · 23-02-2012, 21:15

Quote:

Originally Posted by Sephiroth

I don't know how this discussion has backed itself so far into a corner. Speedtests on the VM network are influenced by:

- Congestion (if any) at the optical node
- Congestion (if any) at the line card
- Congestion (if any) at the CMTS itself (unlikely)
- Congestion (if any) in the core (rare but not unknown)
- Congestion (if any) in the VM backbone (rare and occurs if there is a routing problem elsewhere)
- Congestion (if any) at the peering point (deffo possible)
- Congestion (if any) at the speed test server (e.g. Maidenhead - so well known)

Simples.

so you are saying that the chances are that the congestion is at the line card? due to the number of connections going in to it?

Sephiroth · 23-02-2012, 21:19

In that context, mon General, I'm saying that if there is congeston at the line card it's due to the number of connections going in.

craigj2k12 · 24-02-2012, 01:31

Quote:

Originally Posted by Sephiroth

In that context, mon General, I'm saying that if there is congeston at the line card it's due to the number of connections going in.

I think its more just slow routing, if that possible, pings can still remain low, but the actual capacity is low, the maidenhead server I speak off acheived 500mbit from a server just seconds after getting 30mbit from a VM 100mbit line

Andrewcrawford23 · 24-02-2012, 08:08

Quote:

Originally Posted by craigj2k12

I think its more just slow routing, if that possible, pings can still remain low, but the actual capacity is low, the maidenhead server I speak off acheived 500mbit from a server just seconds after getting 30mbit from a VM 100mbit line

ive seen the maidenhead server achive over 1gbit from janet line it the only one that truly has the capicty

General Maximus · 24-02-2012, 08:35

Quote:

Originally Posted by Sephiroth

In that context, mon General, I'm saying that if there is congeston at the line card it's due to the number of connections going in.

as you said "unlikely" and "rare" for congestion at the cmts and core I thought you was saying that the chances are that if there is any congestion, it will be at the line card. Do the optical nodes run single modem fibre or multi mode?

Sephiroth · 24-02-2012, 08:59

The optical nodes run a fibre pair back to the local hub. As I understand matters, the upstream fibre can be set to use more chunks of bandwidth (6.4 MHz wide) in the upstream frequency plan and then capacity would rise. At the VM end that would be converted to RF and an upstream port is then used at the line card to take the upstream in.

So I believe that the line card, optical node and in some cases the peering points are the major points of congestion.

Does that answer your question?

---------- Post added at 09:59 ---------- Previous post was at 09:52 ----------

Quote:

Originally Posted by craigj2k12

I think its more just slow routing, if that possible, pings can still remain low, but the actual capacity is low, the maidenhead server I speak off acheived 500mbit from a server just seconds after getting 30mbit from a VM 100mbit line

Years of doing this - and empirically the Maidenhead server is notorious for giving low results. Consistently.

Right now, several tests show Maidenhead at 10 meg downstrea; London Namesco shows c. 50 meg.

General Maximus · 24-02-2012, 09:20

yes thanks. The VM network as whole can handle the traffic going through it, the problem is too many people accessing it. It is like having a 10 lane motorway and everyone using a single lane road to get on to it.

qasdfdsaq · 24-02-2012, 13:24

Quote:

Originally Posted by Sephiroth

Years of doing this - and empirically the Maidenhead server is notorious for giving low results. Consistently.

Right now, several tests show Maidenhead at 10 meg downstrea; London Namesco shows c. 50 meg.

On the other hand Namesco gives consistently high pings from just about everywhere. Odd, but I've never bothered investigating it.

craigj2k12 · 24-02-2012, 15:58

Quote:

Originally Posted by Sephiroth

Right now, several tests show Maidenhead at 10 meg downstrea; London Namesco shows c. 50 meg.

I dont think your getting the points, sephy, the maidenhead server has bags of capacity, it can test 500mbit from a server, andrew above has confirmed this by saying it can test gigabit from a janet line, and you have also confimed how a VM 50mbit connection can only reach 10mbit. If the server is capable of gigabit, its clearly a restriction at the ISP end stopping you from getting more than 10mbit

---------- Post added at 16:58 ---------- Previous post was at 16:57 ----------

Quote:

Originally Posted by qasdfdsaq

On the other hand Namesco gives consistently high pings from just about everywhere. Odd, but I've never bothered investigating it.

I think its a highly utilised web server as I have noticed the same

Andrewcrawford23 · 24-02-2012, 17:33

Quote:

Originally Posted by craigj2k12

I dont think your getting the points, sephy, the maidenhead server has bags of capacity, it can test 500mbit from a server, andrew above has confirmed this by saying it can test gigabit from a janet line, and you have also confimed how a VM 50mbit connection can only reach 10mbit. If the server is capable of gigabit, its clearly a restriction at the ISP end stopping you from getting more than 10mbit

---------- Post added at 16:58 ---------- Previous post was at 16:57 ----------

I think its a highly utilised web server as I have noticed the same

i not sure it i restricted i think i have got it give me 20mb give em aminute ill run it jsut now

i am goign to run it from the otehr comptuer which generally get better results

craigj2k12 · 25-02-2012, 00:25

yeah now run it from a high capacity line, or any other high speed non-vm service, you will notice that VM connections give significantly lower results

qasdfdsaq · 25-02-2012, 03:30

Indeed, Maidenhead is playing funny buggers. 4:30 am and still only getting this:

And no it isn't my connection's fault (for once):

Chrysalis · 25-02-2012, 05:42

The #1 reason on VM for speedtests will be congested UBR's, congested transit/peering is probably #2.

Alot of people think VM will have a single fat pipe out to the entire internet and working on that principle if they see they can get full speed from one place they will assume anywhere they dont get full speed from the other end is to blame. Also remembering routing will differ for different VM subnets, so one customer may have poor performance from a particular server whilst another may not. eg speed i/o is horrible on my current ip, but if I change my VM ip to another subnet speed i/o is way faster as the return path changes. For me maidenhead is as fast as namesco, usually if maidenhead slows down then so does namesco and is my UBR port. I actually probably get namesco slowing down more as sometimes the VM core network gets congested on the way to the edge router.

speedtest.net have specific requirements to host a speedtest there, they wouldnt let someone host a speedtest without proper capacity in place. The good advantage of all the different locations is that its a good way to test peering quality of the isp. eg. on easynet I used to get good speeds from everywhere worldwide just about with only a few exceptions, when on murphx it was a very different story which performed ok to places like linx but suffered on a lot of routes. VM is a mixed bag which seems to have good capacity on a few specific routes but poor on others.

craigj2k12 · 25-02-2012, 15:40

Chrys can you run a speed test from one of your servers for a comparison?

and then we can see its a VM issue

Its the same to TBB speed test - always slow from VM connections whereas fine from others. infact, tbb's speedtest is far worse on VM than to maidenhead whereas there seems to be no problem from other ISPs