Re: miivi speed

From: Ty Heath <heath_at_mediadefender.com>
Date: Fri, 18 May 2007 10:54:20 -0700

No, but at least you know that any TCP/IP connection to those clients
should finish quickly. Instead of having the off chance that the
first X clients you connect to are offline, and nothing happening for
X*5 seconds.

I would maybe change the limit to 3 half open connections for the top
most download (most recently queued), and 1 half open connection for
the rest.

Ty

On May 18, 2007, at 10:06 AM, Stefan Kaczmarek wrote:

> Does the UDP thing say whether or not they're busy?
>
> On May 18, 2007, at 10:04 AM, Jay Mairs wrote:
>
>> Right, so wouldn't the UDP ping thing Ty was talking about speed
>> up the connecting process? If you get a list of 100 Ares sources,
>> you ping all of them, then put the ones that respond at the top of
>> the list. Otherwise, it takes almost 10 minutes to try to connect
>> to them all.
>>
>>
>>
>> ----- Original Message -----
>> From: Stefan Kaczmarek <stefan_at_thezonie.org>
>> To: Jay Mairs
>> Cc: Ivan Kwok; Ty Heath; Sergio Alvarez; Dylan Douglas; Ben Ebert;
>> Randy Saaf
>> Sent: Fri May 18 09:54:56 2007
>> Subject: Re: miivi speed
>>
>> Well, like I said, there is a 1 connection per transfer limit that
>> times out every 5 seconds. What usually seems to be happening is
>> that I am connecting to people that are busy, because you see the
>> transfer rate jump up then down just for a sec ... That's what
>> that is.
>>
>> Also, if there are any P2P sources, shit will start transferring
>> right away. And the more P2P sources, the faster it will go.
>>
>> On May 18, 2007, at 9:51 AM, Jay Mairs wrote:
>>
>>
>> Actually, I think the warm-up time may have a bigger
>> effect on the user's perception of speed. Obviously, we would
>> want both to be performing well.
>>
>>
>> ----- Original Message -----
>> From: Stefan Kaczmarek <stefan_at_thezonie.org>
>> To: Ivan Kwok
>> Cc: Ty Heath; Jay Mairs; Sergio Alvarez; Dylan Douglas;
>> Ben Ebert; Randy Saaf
>> Sent: Fri May 18 09:38:40 2007
>> Subject: Re: miivi speed
>>
>> No, there is a limit of 30 connections per transfer, and 1
>> connecting socket per transfer. I also force a time-out if a
>> connecting socket has not connected within 5 seconds.
>>
>> On May 18, 2007, at 1:51 AM, Ivan Kwok wrote:
>>
>>
>> I think both, but the issue of transfer
>> speed is more important. XP SP2 limits the number of half-open TCP
>> connections, not total number of connected TCP connections. As I
>> remember, you made the app to allow total 3 TCP connections only
>> per file, right?
>>
>>
>>
>> -Ivan
>>
>>
>>
>> From: Stefan Kaczmarek [mailto:stefan_at_thezonie.org]
>> Sent: Friday, May 18, 2007 12:02 AM
>> To: Ty Heath
>> Cc: Jay Mairs; Ivan Kwok; Sergio Alvarez; Dylan
>> Douglas; Ben Ebert; Randy Saaf
>> Subject: Re: miivi speed
>>
>>
>>
>> Is this an issue of transfer speed or warm-up time?
>>
>>
>>
>> On May 17, 2007, at 6:00 PM, Ty Heath wrote:
>>
>>
>>
>>
>>
>> Well, considering that the command line version of
>> the app downloads perfectly fine, I think we need to examine the
>> Windows XP SP2 limitations and how Stefan is dealing with those.
>>
>>
>>
>> Some things to speed up the client side:
>>
>>
>>
>> 1. Do a quick UDP strobe with DHT_HELLO_REQ
>> messages to all sources. This will bypass the windows limitations
>> as far as seeing who is online with an open firewall. The client
>> DHT port is the same as the file transfer TCP/IP port.
>>
>>
>>
>> 2. Do custom timeouts for the TCP/IP connections
>> of 3-5 seconds. Ignore the system default timeouts, as these can
>> be stupidly high sometimes.
>>
>>
>>
>> 3. Keep the number of incomplete TCP/IP
>> connections to about 5-6, the maximum total # of connections per
>> download should be upwards of at least 50.
>>
>>
>>
>> 4. Dylan needs to be checking both today's table
>> for sources, and then yesterday's table if today's table isn't
>> returning enough sources.
>>
>>
>>
>> The database is like the live state of the Ares
>> network, so it makes no sense to do another search. Though we may
>> be able to organize the source data differently some how.
>>
>>
>>
>> Ty
>>
>>
>>
>>
>>
>> On May 17, 2007, at 11:32 AM, Jay Mairs wrote:
>>
>>
>>
>>
>>
>> Ty, do you have any thoughts on this?
>>
>>
>>
>> From: Ivan Kwok
>> Sent: Monday, May 14, 2007 6:31 PM
>> To: Jay Mairs; 'Stefan Kaczmarek'; Ty Heath;
>> Sergio Alvarez; Dylan Douglas; Ben Ebert
>> Cc: Randy Saaf
>> Subject: RE: miivi speed
>>
>>
>>
>> Have a real time sources update on demand. When
>> someone request sources of a file, Tyís program will do a Ares
>> DHT sources search in real time and insert the updated results to
>> our DB.
>>
>>
>>
>> -Ivan
>>
>>
>>
>> From: Jay Mairs
>> Sent: Monday, May 14, 2007 6:22 PM
>> To: Stefan Kaczmarek; Ivan Kwok; Ty Heath; Sergio
>> Alvarez; Dylan Douglas; Ben Ebert
>> Cc: Randy Saaf
>> Subject: FW: miivi speed
>>
>>
>>
>>
>>
>> Randy is concerned about the userís perception of
>> the speed of the network and wanted us to discuss it. Below are a
>> few ideas of how to increase network speeds.
>>
>>
>>
>> Health rating refers to a measure or guess about
>> the download speed, probably based on # of sources, existence of
>> thumbnail, etc.
>>
>> Sources could be updated faster or there could be
>> an alternative source of sources.
>>
>> Increasing default upload speed.
>>
>>
>>
>>
>>
>> OK, go
>>
>>
>>
>>
>>
>> From: Randy Saaf
>> Sent: Monday, May 14, 2007 6:06 PM
>> To: Jay Mairs
>> Subject: miivi speed
>>
>>
>>
>> Ideas to increase speed
>>
>>
>>
>> Health rating
>>
>> Getting our sources updated faster
>>
>> Increasing the default upload speeds
>>
>>
>>
>> Anything else?
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>
Received on Fri Sep 14 2007 - 10:55:57 BST

This archive was generated by hypermail 2.2.0 : Sun Sep 16 2007 - 22:19:46 BST