Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 02 Aug 1997 13:47:48 -0700 (PDT)
From:      Simon Shapiro <Shimon@i-Connect.Net>
To:        FreeBSD-SCSI@FreeBSD.ORG
Subject:   DPT Perfromance with Patch 1.1.10
Message-ID:  <XFMail.970802134748.Shimon@i-Connect.Net>

next in thread | raw e-mail | index | archive | help
Some of you have asked or wondered about what performance to expect out of
this release.

Aside from RDBMS testing (which is meaningless for most of us, here are
numbers reported and noted casually on sendero:

*  Sendero is a SuperMicro P6-200 motherboard with 128MB RAM
*  DPT is PM3334UW with 64MB of ECC memory.
*  Two internal disks arranged as RAID-1 are used for /, /usr, etc.  Both
   disks are on Bus 0
*  /usr/src and other active filesystems are on a 2wide RAID-0 in a
   DPT 9W tower using 4GB Baracuda disks.
*  Archives, CVS, etc. are on a 4x RAID-5, in the same cabinet, same type
   disks.
*  All striping values are DPT defaults.  However, the 9W tower is split
   into two busses, and the raid arrays are symetrically balanced across
   the busses.
*  This is a less-than-optimal arrangement for optimal performance, but
   reasonable, considering the resources.
*  A typical job mix is Kernel PPP on a 64K ISDN channel, make world on the
   2.2 tree, building of two kernels simultanously and a large NFS update of
   copper (the next sendero).  All this with email, X11 and other such
   nonsense.  If it makes a difference, the X11 display is on a #9 S3 card,
   running 1600x1200@75Hz.  Yes the type is small on a 17" monitor.

System responsiveness is excellent, no delays in the X11 response, no
crashes in the last 2 weeks or so.

Xprefmon++ reports:

*  780 or more Disk Transfers per second.
*  700 or more NFS Srver Calls/Sec
*  1700 Interrupts/Sec.

We have seen, repeatedly peaks of:
*  1,500 disk i/o per sec
*  900 NFS service calls per sec
*  5,600 (no typo) interrupts/sec.

The high interrupt rate happens when we run the ISDN line at 128Kbps.
The high I/O rate happens when we add stress testing to the mix.

Where we see slow numbers is on the SMP -current kernel.  We are still
working on that one.

The only DPT options in the configuration are:

options     DPT_USE_SINTR=1
options     DPT_MEASURE_PERFORMANCE
options     DPT_COMMAND_SPLHIGH     # Experimental, briefly locks eata ports

Hope this answers some questions.

Simon




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?XFMail.970802134748.Shimon>