Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I am looking into a problem with some images hosted in the US, that some users get really slow load times. This image for example; http://opencirrus-g0805.hpl.hp.com/books40/UMICH/2160/10/47/23/59/216010472359/SCALE/sp3031.jpg

From the US datacentre using apache bench ab;

Server Hostname:        opencirrus-g0805.hpl.hp.com
Document Path:          /books36/IA/2160/40/00/01/15/216040000115/SCALE/sp0831.jpg
Document Length:        62604 bytes
Failed requests:        0
Requests per second:    1666.38 [#/sec] (mean)
Time per request:       6.001 [ms] (mean)
Transfer rate:          102405.91 [Kbytes/sec] received

from another box in the dc;
Requests per second:    1847.39 [#/sec] (mean)
Time per request:       5.413 [ms] (mean)
Transfer rate:          113526.09 [Kbytes/sec] received

From missouri
Requests per second:    27.88 [#/sec] (mean)
Time per request:       358.700 [ms] (mean)
Transfer rate:          1774.29 [Kbytes/sec] received

ping from missouri
PING opencirrus-g0805.hpl.hp.com (198.55.33.68) 56(84) bytes of data.
64 bytes from opencirrus-g0805.hpl.hp.com (198.55.33.68): icmp_seq=1 ttl=51 time=55.9 ms

From London they are getting very slow
Requests per second:    5.29 [#/sec] (mean)
Time per request:       1891.408 [ms] (mean)
Transfer rate:          358.92 [Kbytes/sec] received

even though ping is much less than the time per request;
PING opencirrus-g0805.hpl.hp.com (198.55.33.68) 56(84) bytes of data.
64 bytes from opencirrus-g0805.hpl.hp.com (198.55.33.68): icmp_seq=1 ttl=54 time=177 ms

So for users in London its taking over a second to download a 62K file, and for users about in the US its taking ~650ms

Is that to be expected, and what can I do to speed things up? (as it appears that locally the web-server is able to serve the files very fast to boxes on the LAN - I am not sure what can be done on the web server side)

share|improve this question

2 Answers 2

up vote 1 down vote accepted

From the numbers you've cited I'd say your getting quite good bandwidth - certainly you're never going to get the same bandwidth across continents that you'll get within the same country.

Bear in mind that a HTTP request requires at least 2 and a half round trips (TCP handshake, http request/response, close connection). So the overhead on your US connection is 0.0559*2.5 = 0.14 s, and for UK, 0.177*2.5=0.4425 s

So the time taken to actually deliver the content, including any end point processing for US users is 0.51 seconds , and for the UK, 1.45 seconds.

Or 0.95 MBit/s US and 0.3MBit/s for UK.

How do you speed it up? Well you could use a private network service like Akmai IP accelerator (I expect this will be rather expensive if you want to make the content available on the internet in the UK rather than a branch office setup). Or you could use a content delivery network (CDN).

But before you even attempt this you should make sure that your HTTP caching information is optimized.

share|improve this answer

Don't know if serving content (both dynamic and static) from a local edge would be something you are willing to do, but if you are it's not very difficult for an experienced sysadm to setup a akamai alike caching edge in a geographical place. Some alternatives are FS Big-IP, Bluecoat, or the open source Varnish.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.