Matthew Toseland wrote:
> On Wed, Jul 12, 2006 at 12:46:57PM -0500, David Sowder (Zothar) wrote:
>
>> Matthew Toseland wrote:
>>
>>> On Wed, Jul 12, 2006 at 10:39:49AM -0500, David Sowder (Zothar) wrote:
>>>
>>>
>>>> Matthew Toseland wrote:
>>>>
>>>>
>>>>> On Tue, Jul 11, 2006 at 05:40:22PM -0500, David Sowder (Zothar) wrote:
>>>>>
>>>>>
>>>>>
>>>>>> On my node, which is admittedly not setup to use more than the default
>>>>>> 128 MB of RAM, the MemoryChecker run GC is done roughly every 30
>>>>>> seconds and with the timing and before/after compare I've added
>>>>>> locally, it's taking less than 2 seconds to complete, so this is hardly
>>>>>> a big performance hit, at least on my machine.
>>>>>>
>>>>>> PacketSender perhaps needs to have a minimum sleepTime as well as a
>>>>>> maximum sleep time? That might be one place to reduce CPU load
>>>>>> slightly?
>>>>>>
>>>>>>
>>>>>>
>>>>> Not a good idea IMHO; if it needs to wake up to send a packet, it needs
>>>>> to wake up... they don't become urgent immediately...
>>>>>
>>>>>
>>>>>
>>>> We can't delay by up to 25-50 ms?
>>>>
>>>>
>>> We already delay.
>>>
>>>
>> The theory was that the PacketSender thread wouldn't wait 6 ms for the
>> next scheduled action, but maybe 25 ms to reduce the likelihood that
>> PacketSender is near 100% runnable CPU-wise. I suppose this behavior
>> could be fingerprinted though.
>>
>
> We delay and queue stuff anyway. If we send a packet to a peer, we will
> include any queued messages for that peer; this is packet coalescing.
> Only after some time (200ms iirc) has passed does a message become
> urgent, and it will cause a packet send by itself. We do however wake
> the sender thread up when queueing a packet; maybe we can avoid this.
> I'll have a look sometime, though I doubt it's a major CPU eater.
>
I haven't measured it's affect, but my idea was specifically about the
sleepTime calculation at the end of PacketSender and adding a sleepTime
min of 25 or something in addition to the current sleeptTime Math.max of
200). I implemented this locally, but like I said, I don't have any
measurements. One of these days, I'm probably going to look into
profiling Java.
--
David R. Sowder
Supervisor of Language Acquisition Center
Department of Modern Languages
University of Texas at Arlington
Work: 817-272-5148 davids at uta.edu http://langlab.uta.edu/
Personal: david at sowder.com http://david.sowder.com/
