12:09 a.m.
----- Original Message -----
From: "ralf boogie blues"
<ralf_boogie_blues(a)bluewin.ch>
To: undertow-dev(a)lists.jboss.org
Sent: Monday, 3 November, 2014 2:30:40 AM
Subject: [undertow-dev] ResponseRateLimitingHandler questions
Hi Stuart
Thanks a lot for the ResponseRateLimitingHandler. I played a little
bit with
this handler and do have some questions.
I added some trace logs in order to understand the logic. Of course,
I didn't
understand everything;-) May you will correct me.
First, I tested the download response with a 1 MBytes file. The
response
limit was set to 100000 Bytes/s. This should download the whole file in
about 10s but the result was 18s.
Then, I started to dig into the handler.
How I cReated the handler.
final ResponseRateLimitingHandler responseLimiter = new
ResponseRateLimitingHandler(proxy, 100000, 1, TimeUnit.SECONDS);
The trace logs showed me this (with the settings above):
* the handler writes chunks as big as the buffer is defined, I think
on my
machine 16KBytes.
* it repeats this until the bytes written are greater than the rate limit.
* then, the handler calculates the time to sleep until the next chunks can be
written.
Problem 1: The handler writes always more than the specified limit.
Because
of this, the handler calculates n * units for the time to sleep. So, in my
test the handler slept always 2s instead of 1s.
I think this is a bug, for larger
unit times I should pro-rata the wait time. If you use a smaller step time (e.g. 10k bytes
per 100ms ) this effect will be less pronounced.
Problem 2: The size of the chunks is probably too static. I think the
chunks
should be size relatively to the rate limit.
This can result in lots more write()
calls, which are expensive, so if you are using this because you have a server under heavy
load you could potentially increase the load significantly.
One option would be to cut down the size of the buffer if it is going to go over the
limit, so it will never write more than the specified maximum in the time period.
Stuart
I changed the handler slightly in this way:
* I introduced along to the existing parameters also a chunkSize.
* canSend(): if (byteCount < (bytes - chunks)) { return true; } // it doesn't
write anymore than allowed
* handleWritten(long written): nextSendTime = startTime + time; // instead
of: nextSendTime = startTime + (units * time);
* writeFinal and transferFrom writes not more than bytes <= chunkSize.
This changes gave me the expected result of about 10s download time.
This is quite a long mail :-) It could be still a mistake I did. Let
me know
what you think.
Regards, Ralf
_______________________________________________
undertow-dev mailing list
undertow-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/undertow-dev