Message-Id: <10556.111915@fatcity.com> From: Steve Adams Date: Thu, 13 Jul 2000 12:39:58 +1000 Subject: RE: HP UX-11 I/O vs Sun Solaris 2.6 I/O Hi Tony, The difference is due the different I/O buffering technology that the two operating systems use. HP-UX uses a traditional buffer cache, whereas Solaris does I/O through the virtual memory (paging) subsystem. That means that for dd and C programs, Solaris will get much better cache hit rates than HP-UX, because Solaris can use almost all the physical memory for buffering, whereas HP-UX is limited to half (by default). If you notice the difference from Oracle, it means that db_block_buffers is too small on both systems, but that Solaris is able to compensate because of its paged I/O model whereas HP-UX is not. To compare the relative performance of the I/O subsystems without any distortion from the different buffering techniques, the best you can do is to compare the sum of the average waiting and service times as shown by sar. Also, make sure that your system administrator is not just using the default queue depth of 8 on the HP. It is an old default that should really be increased for modern disks. You can tune it using the scsictl command. Regards, Steve Adams http://www.ixora.com.au/ http://www.oreilly.com/catalog/orinternals/ http://www.christianity.net.au/ -----Original Message----- From: Tony Vecchiet [SMTP:Tony.Vecchiet@xilinx.com] Sent: Thursday, July 13, 2000 11:34 AM To: Multiple recipients of list ORACLE-L Subject: Re: HP UX-11 I/O vs Sun Solaris 2.6 I/O Steve, our sys admin is using the dd command and some custom C programs to read data from various file systems. He repeats all tests on both systems. For 8k blocks, HP is almost 2x slower. On the database side, we're using sql trace and tkprof. Exact same databases, the tkprof output is exactly the same, except the IO time is lots longer. tony Steve Adams wrote: > Hi Tony, > > What statistics are you using to quantify the difference? > > Regards, > Steve Adams > http://www.ixora.com.au/ > http://www.oreilly.com/catalog/orinternals/ > http://www.christianity.net.au/ > > -----Original Message----- > From: Tony Vecchiet [SMTP:Tony.Vecchiet@xilinx.com] > Sent: Thursday, July 13, 2000 8:47 AM > To: Multiple recipients of list ORACLE-L > Subject: HP UX-11 I/O vs Sun Solaris 2.6 I/O > > We are migrating to HP UX-11 from Sun Solaris 2.6 and are > finding I/O to be real slow at the OS level on HP. We're using Oracle 8.0.5.2 > and 8k block sizes (dbblockbuffers). The Sun system uses 7200rpm disks > and ultra scsi, the HP uses their FC60, fiber channel and 10k rpm disks. > On both Sun and HP we are using file systems, not raw partitions. > I'd think that the HP IO system would be faster, but it's much slower (35-50% slower). > > Does anyone have any ideas on what to check or do? > thanks > > << File: tony.vecchiet.vcf >> > -- > Author: Steve Adams > INET: steve.adams@ixora.com.au > > Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051 > San Diego, California -- Public Internet access / Mailing Lists > -------------------------------------------------------------------- > To REMOVE yourself from this mailing list, send an E-Mail message > to: ListGuru@fatcity.com (note EXACT spelling of 'ListGuru') and in > the message BODY, include a line containing: UNSUB ORACLE-L > (or the name of mailing list you want to be removed from). You may > also send the HELP command for other information (like subscribing).