FreeBSD 5.3b7and poor ata performance

[Julian Elischer]

Mike Jakubik wrote:

>Kenneth Culver said:
>
>  
>
>>Quoting fandino <fandino at ng.fadesa.es>:
>>
>>    
>>
>>>Søren Schmidt wrote:
>>>      
>>>
>>>>># dd if=/dev/ad4 of=/dev/null bs=1024k count=1024
>>>>>1024+0 records in
>>>>>1024+0 records out
>>>>>1073741824 bytes transferred in 31.090536 secs (34535970 bytes/sec)
>>>>>          
>>>>>
>>>>etc, and I get this:
>>>>1073741824 bytes transferred in 18.488903 secs (58074934 bytes/sec)
>>>>1073741824 bytes transferred in 14.956484 secs (71791059 bytes/sec)
>>>>        
>>>>
>>ad4: 114440MB <Maxtor 6Y120M0/YAR51EW0> [232514/16/63] at ata2-master
>>SATA150
>>ad6: 114440MB <Maxtor 6Y120M0/YAR51EW0> [232514/16/63] at ata3-master
>>SATA150
>>
>>Set up for striping:
>>
>>ar0: 228880MB <ATA RAID0 array> [29178/255/63] status: READY subdisks:
>>  disk0 READY on ad4 at ata2-master
>>  disk1 READY on ad6 at ata3-master
>>
>>I get the following performance with dd:
>>
>>dd if=/dev/ad4 of=/dev/null bs=1m count=200
>>200+0 records in
>>200+0 records out
>>209715200 bytes transferred in 3.538456 secs (59267433 bytes/sec)
>>
>>dd if=/dev/ad6 of=/dev/null bs=1m count=200
>>200+0 records in
>>200+0 records out
>>209715200 bytes transferred in 3.551439 secs (59050767 bytes/sec)
>>
>>dd if=/dev/ar0 of=/dev/null bs=1m count=200
>>200+0 records in
>>200+0 records out
>>209715200 bytes transferred in 1.837508 secs (114130226 bytes/sec)
>>
>>This is with a custom kernel with all debugging options off. I haven't
>>seen a
>>mail from you yet that says whether you turned off debugging options in
>>the
>>kernel or not, and if you haven't turned off debugging options, I'd
>>strongly
>>suggest doing so.
>>    
>>
>
>Out of curiosity, i ran this on one of our production servers, which runs
>on a dual Xeon MB, with SCSI raid-10 setup, and to my surprise here are
>the results:
>
>CPU: Intel(R) Xeon(TM) CPU 2.80GHz (2799.22-MHz 686-class CPU)
>real memory  = 2146959360 (2047 MB)
>avail memory = 2099650560 (2002 MB)
>FreeBSD/SMP: Multiprocessor System Detected: 2 CPUs
>
>da0 at asr0 bus 0 target 0 lun 0
>da0: <ADAPTEC RAID-10 3B0A> Fixed Direct Access SCSI-2 device
>
>FreeBSD 5.3-BETA4 #0: Sun Sep 12 13:09:43 EDT 2004
>
>(Custom kernel, no debugging)
>
># dd if=/dev/da0 of=/dev/null bs=1m count=200
>200+0 records in
>200+0 records out
>209715200 bytes transferred in 6.225309 secs (33687517 bytes/sec)
>
>Why is a SCSI raid-10 system slower than a plain IDE disk? Something is
>wrong here.
>

I BELIEVE (without empirical proof) that the new scheme of running IO 
through the geom threads
requires a higher degree of smartness from the scheduler than before, 
where IO was done either by interrupt
or by the calling (already running) thread.

Make sure you have preemption enabled, and make sure that HTT is turned 
off. (HTT is a DOG that slows down
almost everything except processes that have a lot of FP work).

>
>
>_______________________________________________
>freebsd-current at freebsd.org mailing list
>http://lists.freebsd.org/mailman/listinfo/freebsd-current
>To unsubscribe, send any mail to "freebsd-current-unsubscribe at freebsd.org"
>  
>