[Spread-users] 1225

koorosh.alahiari at ids.allianz.com koorosh.alahiari at ids.allianz.com
Wed Mar 6 12:39:43 EST 2002


Hi All,

I have done some performance testing & find that
Sending is a little over twice as fast as
Receiving and this is why Rceivers
get shot down by spread since eventually
spread is going to reach its backlog
limit.

My Senders & Receivers (one of each)
do very little else but sending and receiving
in a tight loop.

Does this make sense?

Regards - Koorosh



                                                                                                                                        
                    George Schlossnagle                                                                                                 
                    <george at omniti.com>              To:     koorosh.alahiari at ids.allianz.com                                           
                    Sent by:                         cc:     guido at python.org, spread-users at lists.spread.org,                           
                    spread-users-admin at lists.        spread-users-admin at lists.spread.org                                                
                    spread.org                       Subject:     Re: [Spread-users] 1225                                               
                                                                                                                                        
                                                                                                                                        
                    05/03/02 19:03                                                                                                      
                                                                                                                                        
                                                                                                                                        



> Guess the problem is that messages are accumulating
> in spread too fast. This is a good thing BUT is this feature (of kicking
> off
> humble, slow clients) something that can be prevented or altered
> through configuring spread?

You can configure the maximum backlog at compile time (*cough* soft
limit *cough*), but ultimately, if you always send faster than you
recieve, you will fall behind and not be able to keep up.


>
> Regards, Koorosh
>
>
>
>
>
>                     George Schlossnagle
>                     <george at omniti.com>              To:
> koorosh.alahiari at ids.allianz.com
>                     Sent by:                         cc:
> guido at python.org, spread-users at lists.spread.org,
>                     spread-users-admin at lists.        spread-users-
> admin at lists.spread.org
>                     spread.org                       Subject:     Re:
> [Spread-users] 1225
>
>
>                     05/03/02 17:31
>
>
>
>
>
> It may be due to the delay writing to your tty.  Try a client like
> spreadlogd (http://www.lethargy.org/mod_log_spread/spreadlogd.tar.gz)
> that writes to disk instead (or try one that just joins, reads and
> counts.)
>
>
> On Tuesday, March 5, 2002, at 11:15 AM, koorosh.alahiari at ids.allianz.com
> wrote:
>
>>
>> George,
>>
>> My client (and spuser) & Server are all on the same machine.
>> My undestanding is that TCP/IP is clever enough to detect that
>> and not let the packet leave the machine (I could very well be WRONG
>> here)!
>> And the machine I am trying things on is a 4 processor Sun E10000
>> running
>> Solaris 7.
>>
>> I will try things out with a kind of client that you are suggesting and
>> let
>> you know!
>>
>> Thanks & Best Regards - Koorosh
>>
>>
>>
>>
>>                     George
>>                     Schlossnagle         To:
>> koorosh.alahiari at ids.allianz.com
>>                     <george at omnit        cc:     guido at python.org,
>> spread-users at lists.spread.org,
>>                     i.com>               spread-users-
>> admin at lists.spread.org
>>                                          Subject:     Re:
>> [Spread-users] 1225
>>                     05/03/02
>>                     17:07
>>
>>
>>
>>
>>
>> Ok.... so now spuser isn't processing fast enough.   Is there a high
>> degree of latency or packet-loss on your network?  Have you tried a
>> client which does not update the screen for every message?
>>
>>
>> On Tuesday, March 5, 2002, at 10:58 AM,
>> koorosh.alahiari at ids.allianz.com
>> wrote:
>>
>>>
>>> Not joining the group got me a bit further (up to 7419 see below).
>>> After that spuser was kicked off! BUT my client thinks it has
>>> delivered
>>> all 100,000 messages.
>>>
>>> User>
>>> ============================
>>> received RELIABLE message from #XXXXXXXX, of type 0, (endian 0) to 1
>>> groups
>>> (17 bytes): Test message 7419
>>>
>>> User>
>>> ============================
>>> SP_error: (-8) Connection closed by spread
>>>
>>> ============================
>>>
>>> Bye.
>>>
>>>
>>>
>>>
>>>                     George Schlossnagle
>>>                     <george at omniti.com>              To:     Guido van
>>> Rossum <guido at python.org>
>>>                     Sent by:                         cc:
>>> koorosh.alahiari at ids.allianz.com, spread-users at lists.spread.org
>>>                     spread-users-admin at lists.        Subject:     Re:
>>> [Spread-users] 1225
>>>                     spread.org
>>>
>>>
>>>                     05/03/02 16:44
>>>
>>>
>>>
>>>
>>>
>>> To elaborate (perhaps superfluously).  Your sender need not sp_join()
>>> the group it is sp_multicast()'ing to.  In fact, unless you have it
>>> setup to read messages at well, you do not want ti to join the group,
>>> otherwise you will get the behavior you are seeing here.
>>>
>>> George
>>>
>>> On Tuesday, March 5, 2002, at 10:41 AM, Guido van Rossum wrote:
>>>
>>>>> I have a client that generates the messages and I run spuser as
>>>>> another
>>>>> client
>>>>> that joins the group my program sends the messages to.
>>>>> spuser printsout all the messages (up to 1225) BUT my program
>>>>> does not try the to receive the messages that it is
>>>>> sending itself.
>>>>> Looks like things are happening too fast
>>>>> for spread to handle correctly!
>>>>
>>>> Is the sender a member of the group to which it sends the messages?
>>>> Then you're running into the 1000-message limit -- I bet the sender
>>>> isn't set up to receive its own messages, but the Spread semantics
>>>> make this happen.
>>>>
>>>> --Guido van Rossum (home page: http://www.python.org/~guido/)
>>>>
>>>>
>>>> _______________________________________________
>>>> Spread-users mailing list
>>>> Spread-users at lists.spread.org
>>>> http://lists.spread.org/mailman/listinfo/spread-users
>>>>
>>>>
>>> // George Schlossnagle
>>> // Principal Consultant
>>> // OmniTI, Inc                      http://www.omniti.com
>>> // (c) 301.343.6422   (e) george at omniti.com
>>> // 1024D/1100A5A0  1370 F70A 9365 96C9 2F5E 56C2 B2B9 262F 1100 A5A0
>>>
>>>
>>>
>>> _______________________________________________
>>> Spread-users mailing list
>>> Spread-users at lists.spread.org
>>> http://lists.spread.org/mailman/listinfo/spread-users
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>> // George Schlossnagle
>> // Principal Consultant
>> // OmniTI, Inc                      http://www.omniti.com
>> // (c) 301.343.6422   (e) george at omniti.com
>> // 1024D/1100A5A0  1370 F70A 9365 96C9 2F5E 56C2 B2B9 262F 1100 A5A0
>>
>>
>>
>>
>>
>>
>>
> // George Schlossnagle
> // Principal Consultant
> // OmniTI, Inc                      http://www.omniti.com
> // (c) 301.343.6422   (e) george at omniti.com
> // 1024D/1100A5A0  1370 F70A 9365 96C9 2F5E 56C2 B2B9 262F 1100 A5A0
>
>
>
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users
>
>
>
>
>
>
>
// George Schlossnagle
// Principal Consultant
// OmniTI, Inc                      http://www.omniti.com
// (c) 301.343.6422   (e) george at omniti.com
// 1024D/1100A5A0  1370 F70A 9365 96C9 2F5E 56C2 B2B9 262F 1100 A5A0



_______________________________________________
Spread-users mailing list
Spread-users at lists.spread.org
http://lists.spread.org/mailman/listinfo/spread-users










More information about the Spread-users mailing list