Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 02 Nov 1998 01:09:43 -0500 (EST)
From:      Simon Shapiro <shimon@simon-shapiro.org>
To:        Simon Shapiro <shimon@simon-shapiro.org>
Cc:        freebsd-alpha@FreeBSD.ORG
Subject:   RE: Load Test in Progress
Message-ID:  <XFMail.981102010943.shimon@simon-shapiro.org>
In-Reply-To: <XFMail.981101193230.shimon@simon-shapiro.org>

next in thread | previous in thread | raw e-mail | index | archive | help

Simon Shapiro, On 02-Nov-98 you wrote:
>  I am running, overnight, a load test on the PW 433au.  It is equipped
>  with
>  512MB of RAM and 2MB of L2 (L3?) cache.
>  
>  The test consists of 512 instances of st.c, all against a 128MB
>  partition
>  on the only disk in the system.
>  
>  It started with LA climbing to about 220.xx, and gradually declining to
>  7.84 (?!).  The funny thing is that there is considerable physical I/O
>  going on.
>  
>  St.c simply seeks at random in the file/device given and reads 4KB from
>  it
>  (it knows to do many other evil things).
>  
>  I would have thought that FreeBSD kernel would have completely cached
>  the
>  128MB.
>  
>  Response to the keyboard and network stays reasonably good, considering
>  the
>  load.  A 2xP6-200 machine typically goes on vacation at this load.
>  
>  So, this looks real good, but some questions remain.  I will leave the
>  test
>  running overnight.  If it is still alive, I'll add some nasty netwok
>  traffic to it.
>  
>  Great Job! and thanx for all this ongoing help!
>  
>  Simon

More interesting observations:

*  I NFS mounted /usr/src from nomis (2xP6-200) and ran 5 parallel find |
   cpio sets, dumping to /dev/null.

*  The NFS activity peaked and held at almost 4,100 packets/sec each way
   (on nomis, no way to measure on the Alpha yet) and about 370 NFS
   transfers/sec. If NFS transfers are 4K, then we have 1.5MB/Sec.  This is
   over 100MBit Ethernet.  I know this is comparable to a medium size,
   older generation NetApps, but is it good?

*  The LA busieness is baffling.  While this load seemed to peak it at 220
   or so, it slowly went down (same load!) to 0.6, only to oscilate back up
   to 15.xx. or even 30.xx.  This is most bizarre.  Responsiveness seemed
   to track these numbers so they are somehow anchored in some reality. 
   But what does it all mean?

*  Later inspection shoed on the console a loop of bash sessions dying on
   sig 11.  Maybe they are associated with the daily crons that decided to
   kick in just before midnight?  In any case, they ceased once I killed
   the 512 st.c that were running.  Then there was a flurry of inetd deaths
   and sshd death.  Maybe some system resource exhausted?

Overall, the system sustained this load very well.  I'd say, that in this
test it does as well as, if not better than the i386.  Most encouraging!

Simon



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-alpha" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?XFMail.981102010943.shimon>