Hi All,
I have good news! Here's our setup:
SAN Storage Node: E-250 x 2 GB x 2-400 MHz, Solaris 8, JNI FCX-6462 (2Gb) w/
v.5.2.1 driver , 2
STK 9840A
SAN Switch: Brocade Disk: Hitachi 9970, Single ~500 GB UFS Filesystem
w/ Oracle "stuff"
The problem was that a single save stream was only producing on avg about 7
MB/Sec when we knew that a single 9840A could easily write 25-30+ MB/Sec. We
subsequenly split out the above filesystem into individual save sets for each
subdir. This greatly improved things but when it got down to the actual data
files only two save streams were left - each doing a pathetic job.
We know that having 1 HBA for disk and tape not a best practice and it is
planned to install an additional HBA as soon as we get some LC/LC fibre in.
However, we still needed to make it work.
On the HBA, persistent binding was not being used so that was configured after
the tape drives swapped around on us. Below is our jnic146x.conf:
jnic146x0-FcLoopEnabled=0;
jnic146x0-FcFabricEnabled=1;
jnic146x0-FcLinkSpeed=3;
jnic146x0-JniCreationDelay=5;
jnic146x0-FlogiRetryCount=3;
jnic146x0-PlogiRetryCount=5;
jnic146x0-PlogiControlSeconds=30;
jnic146x0-SwitchGidPtSyncEnable=0;
jnic146x0-PlogiRetryTime=50;
jnic146x0-BusyRetryDelay=5000;
jnic146x0-FailoverDelay=120;
jnic146x0-TimeoutResetEnable=0;
jnic146x0-QfullRetryCount=5;
jnic146x0-QfullRetryDelay=5000;
jnic146x0-LunRecoveryInterval=50;
jnic146x0-LunThrottle=64;
jnic146x0-LunDiscoveryMethod=0;
jnic146x0-CmdTaskAttr=1;
jnic146x0-OverrunFailoverCount=0;
jnic146x0-FcIpEnable=1;
jnic146x0-FcFlogiTimeout=10;
jnic146x0-FcLinkUpRecoveryTime=1000;
jnic146x0-FcEngHeartbeatInterval=5;
jnic146x0-def_hba_binding="null";
def_hba_binding="null";
#
jnic146x0-def_wwnn_binding="xxxxxxxxxxxxxxxx";
jnic146x0-def_wwpn_binding="xxxxxxxxxxxxxxxx";
jnic146x0-def_port_binding="$xxxxxx";
#
# 9840
#
jnic146x0-target0_port="$0112e8";
jnic146x0-target0_lun0_hba="jnic146x0";
jnic146x0-target0_lun0_throttle=1;
#
# 9840
#
jnic146x0-target1_port="$0111e8";
jnic146x0-target1_lun0_hba="jnic146x0";
jnic146x0-target1_lun0_throttle=1;
#
# 9970
#
jnic146x0-target2_port="$021100";
jnic146x0-target2_hba="jnic146x0";
jnic146x0-target2_throttle=64;
Next, we tuned the box to maximize I/O regardless of system impact. The 250 is
a dedicated system that ONLY performs backups.
/etc/system:
* Begin Legato NetWorker Tuning
* set maxpgio = 1024
set maxphys = 8388608
* End Legato NetWorker Tuning
*
* Begin 9970 settings
set sd:sd_io_time=0x3c
* End 9970 settings
We also set the maxcontig to 128 (using 'tunefs -a') for the filesystem before
mounting it. Caution: Be sure to read Sun Alert Notification #23429 for
'maxcontig' settings for SCSI and IDE controllers.
** The results:
We see a solid 25-35 MB/Sec for a single stream w/ decent sized files (>1+ MB).
When many small files are being processed the thoughput would vary anywhere
between 8-20 MB/Sec; However, during this time the 'r/s' (reads per second) of
the filestem was between 500-800. This is easy a 3X improvement.
Using two streams in parallel we see 40-50 MB/Sec total throughput and at this
point the 9970 starts to become the bottle neck as the '%busy' field of iostat
ranges between 90-100%.
%iowait (according to top) avgs between 40-50%, 'save' consumes 10-20% CPU, and
15 minute load avg is currently at 0.61 and the backup has been running for a
couple of hours now.
Overall, I'm very satisfied with this performance given the h/w that we are
using.
Hopefully this will help anybody out that is experiencing this type of
performcne problems.
I received a couple of responses but nothing solid as far as tuning guidelines.
Regards,
/\/elson
--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
|
