Amazon Web Services Blog

I have been experimenting with S3 but find that there is a great deal of latency when requesting _known_ objects.

In my simplest test case, requesting a known object from a bucket which contains only two objects takes on average 180-220ms.  Note: this is not counting the time spent creating the request or sending the request but just the time that it takes for a PHP socket_read to return any data.  Requesting an object which does not exist takes on average 60-70ms.  Example code:

$obj = new s3();
$obj->connect();
$obj->requestObject($bucket, $key);    // generates a request and writes to the socket
$timer->start();
$data = $obj->readObject();    // reads response from the socket
error_log(sprintf("took %s sec", $timer->get_time()));

The readObject() function merely calls socket_read with a maximum number of bytes to return set to 4096.  The $key requested in the test case is not even this large.  Also, readObject() returns raw data and does not parse the response header.

In the following test case, the response time is rather quick leading me to believe that S3 just takes a while to actually locate a known object within S3:

$obj = new s3();
$obj->connect();
$obj->requestObject($bucket, $key);    // generates a request and writes to the socket
usleep(2000000);    // sleep for two seconds
$timer->start();
$data = $obj->readObject();    // reads response from the socket
error_log(sprintf("took %s sec", $timer->get_time()));

I am hoping to stream media from S3, but this delay is very noticeable when scanning through clips.  Will I be able to leverage S3 into my application?  Or will there always be a long delay while S3 locates my files on their storage network before sending me back some data?

Thanks in advance!

-Abe