[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-ha-dev
Subject:    Re: [Linux-ha-dev] Re: hearbeat v1.0.2 on Alpha?
From:       Alan Robertson <alanr () unix ! sh>
Date:       2003-06-18 17:12:01
[Download RAW message or body]

Thomas Heiligenmann wrote:
> Alan Robertson wrote:
> 
>>
>> Wow!  That's way better.
>>
>> It still seems to me there are some inaccuracies that should be 
>> corrected...  Let me think this through...
>>
>> Let's say we're trying to time 1.5 seconds (1500ms), and Lc_Hz is 1024...
>>
>> Then this code would give us an answer of 1024...  But the answer 
>> should be
>> 1024+512, which is 1024 + (msec * Lc_Hz) / 1000.  I originally 
>> computed the ms/tick thing primarily to avoid this extra 
>> multiplication - because I saw Lc_Hz/1000 was a constant.
>>
>> But this formula is more accurate as hz gets larger and when 1000 
>> isn't evenly divisible by Hz...  So, I think maybe I'll drop the 
>> "optimization".
>>
>> What if the line were replaced with:
>>
>>     result = secs * Lc_Hz + (msec * Lc_Hz)/1000
>>
>> And for most modern processors, an extra multiply is faster than a 
>> branch...
>>
>> Wouldn't that do it?
>>
> 
> Yes. The formula is kept unchanged and there's no more danger of a 
> division by zero.
> 
>> Would you test this version for me?
>>
> 
> It works with 1.0.2 and 1.1.1 and timing looks fine. Are there any tests 
> I can do to see if there's an improvement in accuracy?
> 
> BTW in 1.1.1 my compiler complained about ipcsocket.c and I had to add 
> an explicit typecast. The whole new diff is attached.
> 
> 
> Thomas
> 
> 
> ------------------------------------------------------------------------
> 
> diff -urP heartbeat-1.1.1/lib/clplumbing/ipcsocket.c heartbeat-1.1.1-patched/lib/clplumbing/ipcsocket.c
> --- heartbeat-1.1.1/lib/clplumbing/ipcsocket.c	Sat May  3 15:27:48 2003
> +++ heartbeat-1.1.1-patched/lib/clplumbing/ipcsocket.c	Wed Jun 18 14:11:04 2003
> @@ -795,7 +795,7 @@
>  
>  		if (sendrc != msg->msg_len) {
>  			cl_perror("Sent %d byte message body: rc = %d"
> -			,	msg->msg_len, sendrc);
> +			,	(int)msg->msg_len, sendrc);
>  		}
>  
>  #if 0
> diff -urP heartbeat-1.1.1/lib/clplumbing/longclock.c heartbeat-1.1.1-patched/lib/clplumbing/longclock.c
> --- heartbeat-1.1.1/lib/clplumbing/longclock.c	Mon Apr 28 18:12:07 2003
> +++ heartbeat-1.1.1-patched/lib/clplumbing/longclock.c	Wed Jun 18 14:06:32 2003
> @@ -109,7 +109,7 @@
>  	if (ms == 0) {
>  		return (longclock_t)0UL;
>  	}
> -	result = secs * Lc_Hz + (msec/ms_per_tick);
> +	result = secs * Lc_Hz + (msec * Lc_Hz)/1000;
>  
>  	if (result == 0) {
>  		result = 1;


I put these into CVS -- along with removing the ms_per_tick variable.


-- 
     Alan Robertson <alanr@unix.sh>

"Openness is the foundation and preservative of friendship...  Let me claim 
from you at all times your undisguised opinions." - William Wilberforce

_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.community.tummy.com
http://lists.community.tummy.com/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
[prev in list] [next in list] [prev in thread] [next in thread]