Here's the same file (91M, quicktime compressed .mov) served in different locations:

1) Fremont, CA: http://fremont30.linode.com/Tailing-Aaron.mov
2) Dallas, TX: http://www.theshore.net/moto/20060825_S ... -Aaron.mov (served by a Linode)
3) Atlanta, GA: http://atlanta1.linode.com/Tailing-Aaron.mov
4) Newark, NJ: http://newark1.linode.com/Tailing-Aaron.mov

I, for one, can max out my cable modem from Nashville (about 800k sec on a good day).

EDIT: Please use http://www.linode.com/speedtest/

-Chris

thx!

I have 15 megabit (downstream) Verizon FIOS, in the Dallas area.

My download speeds for the test files:

• Fremont: 1680 Kbytes/sec
• Dallas: 1890 Kbytes/sec
• Atlanta: 1140 Kbytes/sec

However, I actually have lower ping times to the Atlanta host, compared to the Dallas host. A few weeks ago, I noticed that the hop from alter.net in DFW to theplanet's alter.net gateway (also in DFW) added 35 ms. I think something is messed up in the routing: you can practically go all the way to Chicago and back in that time. But, I don't know who to report it to....

Here's two sets of results.

I tested these sets twice, on Saturday March 3rd and Sunday March 4th 2007 at between 07:00 and 09:00 UTC and got very similar results.

Numbers are ping time; max rate; avg rate.

Download to Orange 20Mbit/sec ADSL line in The Netherlands (NL)

• Linode.com/Dallas ... 129ms; 1.6Mbyte/sec; 1.09Mbyte/sec
• Linode.com/Fremont ... 177ms; 620Kb/sec; 502Kbyte/sec
• Linode.com/Atlanta ... 133ms; 497Kb/sec; 485Kbyte/sec
• Internap/Atlanta ... 120ms; 575Kbyte/sec; 552Kbyte/sec
• MyLinode/Atlanta ... 133ms; 255Kbyte/sec; 250Kbyte/sec

Download to MyLinode/Atlanta

• Internap/Atlanta ... 16ms; 1.12Mbyte/sec; 1.11Mbyte/sec
• Linode/Atlanta ... 0.6ms; 1.08Mbyte/sec; 1.02Mbyte/sec

Notes:
Internap = ftp-mirror.internap.com (41M file)
MyLinode = My Atlanta-based Linode on host77, on the same host machine as the Atlanta download test.


Summary:

• Highest transfer rate on MyLinode is 1.12Mbyte/sec when downloading from a nearby Debian mirror.
• Linode/Dallas provides the highest download rate to AMSIX (Amsterdam); Linode/Atlanta the most consistent.
• When downloading from MyLinode to NL, the max rate is only half that of the host. Is this true for all Linodes?
• In download tests from previous MyLinode/Fremont (host19), the max rate was always ~200Kbyte/sec. [Feb 2007]

Cliff

'Nuff said.

Not really.

I guess though your Linode is at Dallas DC, simply because downloads to Linode at Atlanta (and Fremont, in my experience) never reach that rate.

Here's your test repeated from Atlanta (host77):

linode-atlanta$ wget http://cachefly.cachefly.net/10mb.test
--09:15:53--  http://cachefly.cachefly.net/10mb.test
           => `10mb.test'
Resolving cachefly.cachefly.net... 205.234.175.175
Connecting to cachefly.cachefly.net|205.234.175.175|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10,485,760 (10M) [application/octet-stream]

100%[==========================================================>] 10,485,760     1.08M/s    ETA 00:00

09:16:02 (1.07 MB/s) - `10mb.test' saved [10485760/10485760]

This fits with tests I ran last week (above).

Cliff

From my Linode in Fremont - which is on host56, the Xen beta machine:

peter@fremont ~ $  wget http://cachefly.cachefly.net/10mb.test
--18:21:16--  http://cachefly.cachefly.net/10mb.test
           => `10mb.test'
Resolving cachefly.cachefly.net... 205.234.175.175
Connecting to cachefly.cachefly.net|205.234.175.175|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10,485,760 (10M) [application/octet-stream]
 
100%[====================================>] 10,485,760    10.61M/s             
 
18:21:17 (10.61 MB/s) - `10mb.test' saved [10485760/10485760]
 
peter@fremont ~ $

_________________
/ Peter

Both of my Linodes are at the Dallas DC, and here are my results:

Linode 128 (will become Linode 256 when the upgrades come to Dallas) on host37:

kimandre@edison ~ $ wget http://cachefly.cachefly.net/10mb.test
--14:33:10--  http://cachefly.cachefly.net/10mb.test
           => `10mb.test'
Resolving cachefly.cachefly.net... 205.234.175.175
Connecting to cachefly.cachefly.net|205.234.175.175|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10,485,760 (10M) [application/octet-stream]

100%[================================================================================>] 10,485,760     2.10M/s    ETA 00:00

14:33:15 (1.97 MB/s) - `10mb.test' saved [10485760/10485760]

And Linode 256 (will become Linode 512 when the new upgrades arrive) on host43:

kimandre@newton:~$ wget http://cachefly.cachefly.net/10mb.test
--14:32:23--  http://cachefly.cachefly.net/10mb.test
           => `10mb.test'
Resolving cachefly.cachefly.net... 205.234.175.175
Connecting to cachefly.cachefly.net[205.234.175.175]:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10,485,760 [application/octet-stream]

100%[================================================================================>] 10,485,760     2.08M/s    ETA 00:00

14:32:28 (1.93 MB/s) - `10mb.test' saved [10485760/10485760]

You'd be wrong, I'm in fremont. From dallas though:
'


Course it's only 6 hops from Fremont, and 9 from Planet. It is in Equinix though which has a fair chunk of connectivity.

7 hops from Atlanta (pingtime ~17ms).

In summary then, the download rates (from http://cachefly.cachefly.net/ to various Linodes) are:

Atlanta (UML/cliff)             ...   1.1M/s
Fremont (xen/pclissold)         ...  10.6M/s
Fremont (UML/overlordq)         ...   3.41M/s
Dallas (UML/NeonNero,overlordq) ...  ~2.10M/s

FWIW, from a UML linode on host21 at HE I got

% wget http://cachefly.cachefly.net/10mb.test
--06:47:05--  http://cachefly.cachefly.net/10mb.test
           => `10mb.test'
Resolving cachefly.cachefly.net... 205.234.175.175
Connecting to cachefly.cachefly.net|205.234.175.175|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10,485,760 (10M) [application/octet-stream]

100%[====================================>] 10,485,760     1.52M/s    ETA 00:00

06:47:11 (1.94 MB/s) - `10mb.test' saved [10485760/10485760]

_________________
Rgds
Stephen
(Linux user since kernel version 0.11)

Since this is in the performance and tuning section, I might recommend tuning your tcp settings in the kernel


Here's from Fremont after some tomfoolery:

ovrlrdq@thedarkcitadel:~$ wget http://cachefly.cachefly.net/10mb.test
--00:57:45--  http://cachefly.cachefly.net/10mb.test
           => `10mb.test'
Resolving cachefly.cachefly.net... 205.234.175.175
Connecting to cachefly.cachefly.net|205.234.175.175|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 10,485,760 (10M) [application/octet-stream]

100%[============================================================>] 10,485,760     9.88M/s

00:57:47 (9.87 MB/s) - `10mb.test' saved [10485760/10485760]

Course congestion at the time could have been a factor.

Please share your tomfoolery tips so the rest of us might enjoy a 3-10x speed boost ... .

(Congestion is an unlikely cause ... my measurements were taken at random points on Saturday and Sunday mornings (UTC).)

Cliff

http://dsd.lbl.gov/TCP-tuning/linux.html

YMMV of course

Thanks for this reference.

Trying it out:

BEFORE:
    net.core.rmem_max=109568
    net.core.wmem_max=109568
    net.ipv4.tcp_rmem='4096 87380 87380'
    net.ipv4.tcp_wmem='4096 16384 65536'

    ifconfig eth0 txqueuelen 1000

AFTER:
    net.core.rmem_max=16777216
    net.core.wmem_max=16777216
    net.ipv4.tcp_rmem='4096 87380 16777216'
    net.ipv4.tcp_wmem='4096 65536 16777216'

    ifconfig eth0 txqueuelen 1000

Unfortunately it made no difference at all -- I still max out at ~1.1M/s from both cachefly and host77 (my Linode's host machine).

Cliff