All Read Please: Geographical Request Latency

[Martyn]

Hi there,

I am investigating request latency for AppEngine after experiencing
significantly lower performance then reported in the Java status
results.

This appears to be due to network topology, but I thought it would be
useful to get results from several locations worldwide.

To that end I have created a simple version of an app I am working on,
that provides request timings. I would really appreciate it if people
could take a minute, try the link out and report back on the results.

Here is the link: http://performance.latest.pet-software.appspot.com

many thanks in advance

- Martyn

[Martyn]

I'll start off by reporting that:

In South-East England I am seeing typical latency reported at ~170ms,
but generally ranging between 160ms and 220ms

[Joshua Smith]

Near Boston: low 84ms, high 400ms, median around 95ms.

[Koen Bok]

Amsterdam: ~150ms low: 130ms, high 240ms

[bFlood]

Somerset, NJ: low 80ms, high 340ms, median around 90ms.

[Jarek Zgoda]

Warsaw, Poland here. Typical ~240ms, ranging from 210ms to 740ms (!).

[Martyn]

Ouch!

[Takashi Matsuo]

Tokyo Japan: 220ms ~ 300ms

--
Takashi Matsuo

[Scott]

Melbourne, Australia:

low: 352ms, high 375ms, average approx 360ms

[Albert]

1711 ~ 4632ms

Philippines

[dom.eav]

Phnom Penh, Cambodia: typical latency is 500ms. Very consistent, 487ms
~ 536ms.

[Rodrigo Moraes]

São Paulo, Brazil - average 210ms

-- rodrigo

[Federico Builes]

Medellín, Colombia - Avg. 157ms.

[hawkett]

Orange, NSW, Aust -

low: 380ms, high: 480ms, avg: 440ms

Also moved from UK to Australia recently, and anecdotal 'sluggishness'
definately more noticeable in Aus than in the UK - Melbourne, Sydney &
Orange.

[hawkett]

Also from Orange, Australia -

ping performance.latest.pet-software.appspot.com
PING appspot.l.google.com (66.102.11.141): 56 data bytes
64 bytes from 66.102.11.141: icmp_seq=0 ttl=57 time=20.131 ms
64 bytes from 66.102.11.141: icmp_seq=1 ttl=57 time=24.465 ms
64 bytes from 66.102.11.141: icmp_seq=2 ttl=57 time=20.498 ms
64 bytes from 66.102.11.141: icmp_seq=3 ttl=57 time=20.675 ms
64 bytes from 66.102.11.141: icmp_seq=4 ttl=57 time=22.383 ms
64 bytes from 66.102.11.141: icmp_seq=5 ttl=57 time=22.315 ms

So there's not too much wrong with the underlying infrastructure.

[Ivan Zuzak]

Location: Zagreb, Croatia
Low - 167, High - 195, Avg - ~175

Cheers,
Ivan

[Martyn]

That's interesting!

[Martyn]

I repeated Hawketts ping results for performance.latest.pet-
software.appspot.com and got similar times ~22ms.

So can anyone suggest why we experience such diverse request latency
if the underlying infrastructure is so good? - Hawkett (Australia)
reported avg 440ms with a ping of 22ms and I get 160ms with similar
ping response.

Something else is happening. Note that my request logs report process
time consistently around 23ms.

[Martyn]

Okay, just spoke with a network expert.

He suggests that Google are "any-casting" the DNS. So the pet-
software ping is not really doing a round-trip to the app-server.
Where "perhaps" the appspot.com ping is.

[ajacks504]

New Orleans, LA : 170 - 185 ms

[Ben Nevile]

Am I stupid, or does this test depend as much or more on the "last
mile" and its transport as it does on the underlying network
topology? I'm in Vancouver, Canada, and I'm getting results between
300ms and 600ms, but I'm also using a cellular modem right now.

Ben

[Martyn]

Hi Ben,

Clearly all parts of the network add latency. The bit that adds the
most is the most significant - wherever it may be. The claimed
AppEngine latency in the status figures (for Java and Python) is based
on a round-trip http request, I guess that is not necessarily a
browser test. Hopefully, it is not an http request from somewhere
else on Googles network.

But no, any one figure cannot indicate a problem with app engine
responsiveness, which is why I am looking for a number of responses
worldwide to provide some useful info.

I've yet to see anyone outside the states to report a sub 100ms
response.

In your case, your cellular modem may be responsible for a hefty
slice. It would be interesting if you could also plug into a decent
network and do the same test.

- Martyn

[Mariano Benitez]

Buenos Aires, ~210ms no peaks

[Diego Medina]

US North Carolina: First request 130ms, then ~65ms - 75ms (Cable modem
here)

Diego

[GregF]

Queenstown New Zealand: 440-490ms.

[gjs]

Home Ajax request latency of 425ms
Page One Ajax request latency of 439ms
Page Two Ajax request latency of 420ms
More Ajax request latency of 836ms

After trying multiple times most responses are around 380~420ms

From Australia ( with chrome on windoze xp )

Regards


On Aug 17, 9:02 am, GregF <g.fawc...@gmail.com> wrote:
> Queenstown New Zealand: 440-490ms.

[Ben Nevile]

Plugged in to DSL I get ~150ms.

Ben

[Juraj Vitko]

Prague CZ, O2 ADSL

Ajax request latency of 268ms

(this is when I hit the 'Go' button again - the first one was ~700ms -
but the page's HTML was actually loading at that time I think)

[molicule]

Location: Los Altos, CA (10 mins from google-hq)
average latency: 170ms

If you are planning on collating/summarizing all this info, i'd
appreciate it if you
could send out a link to the summary.

- S.. Sriram

[Hrishi]

77ms to 127ms

Rutgers University, New Jersey, US

[hawkett]

Ok, well I'm now seeing much *worse* (~600ms) latency in Sydney -
reliably slower since the maintenance last night.

ping performance.latest.pet-software.appspot.com
PING appspot.l.google.com (66.102.11.141): 56 data bytes

64 bytes from 66.102.11.141: icmp_seq=0 ttl=54 time=18.807 ms
64 bytes from 66.102.11.141: icmp_seq=2 ttl=54 time=23.454 ms
64 bytes from 66.102.11.141: icmp_seq=3 ttl=54 time=15.535 ms
64 bytes from 66.102.11.141: icmp_seq=4 ttl=54 time=12.386 ms
64 bytes from 66.102.11.141: icmp_seq=5 ttl=54 time=26.489 ms
64 bytes from 66.102.11.141: icmp_seq=6 ttl=54 time=13.917 ms

[gjs]

And -

Pinging appspot.l.google.com [66.102.11.141] with 32 bytes of data:

Reply from 66.102.11.141: bytes=32 time=36ms TTL=52
Reply from 66.102.11.141: bytes=32 time=36ms TTL=52
Reply from 66.102.11.141: bytes=32 time=36ms TTL=52
Reply from 66.102.11.141: bytes=32 time=36ms TTL=52

Ping statistics for 66.102.11.141:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 36ms, Maximum = 36ms, Average = 36ms

Regards