Operating System - HP-UX
1747982 Members
4622 Online
108756 Solutions
New Discussion юеВ

Re: Sybase 12.5, HP-UX 11.11 Kernel ... Tuning ...

 
A. Daniel King_1
Super Advisor

Sybase 12.5, HP-UX 11.11 Kernel ... Tuning ...

Several points:

(1) I'd like to follow-up on:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=183795&admit=-682735245+1156851055506+28353475

Regarding /dev/async. I understand that the minor number can be set to 256 (2^8) under HP-UX 11.0. Does this carry over to v 11.11? What does this bit do under 11.11? Does it still, "defer memory locking?" Is this advisable under newer versions of the OS/DB (since the referenced posting)?

(2) Will this conflict with (or complement) the minor number bits:

0x1 - Don't disable immediate reporting.
0x2 - "Shortens the path of some calls and can improve performance."
0x4 - Enable disk timeouts.

(3) For production, is there a reason all four of these bits should not be used?

(4) So far, I've seen information on four bits, 0, 1, 2 and 8, for a total minor number of 263. Are there other bits which might enhance performance of Sybase using raw devices?

(5) While I'm at it ... what HP-UX kernel parameters relate to the Sybase parameters below:

(a) max async i/os per engine
(b) disk i/o structures
(c) max async i/os per server

There is opportunity for several 10-pointers here! Hey, is Dietmar around?
Command-Line Junkie
5 REPLIES 5
Steven E. Protter
Exalted Contributor

Re: Sybase 12.5, HP-UX 11.11 Kernel ... Tuning ...

Shalom,

Good kernel tuning from 11.00 carries over to 11.11.

vx_ninode # set to 90% of ninode
ninode # set to a reasonably high level to avoid i/o problems.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
A. Daniel King_1
Super Advisor

Re: Sybase 12.5, HP-UX 11.11 Kernel ... Tuning ...

Good point, Protter. The system in question is not only a Sybase sever, but also has many file systems.

I'm uncertain, though, about the ramifications in a mostly-vxfs system:

- vx_ninode set to 0 should be automatic according to physical memory, right?

- ninode is only for hfs, right?
Command-Line Junkie
A. Daniel King_1
Super Advisor

Re: Sybase 12.5, HP-UX 11.11 Kernel ... Tuning ...

http://www.sybase.com/detail?id=1010308

This tuning guide (IE Only!) looks a bit dated ...
Command-Line Junkie
A. Daniel King_1
Super Advisor

Re: Sybase 12.5, HP-UX 11.11 Kernel ... Tuning ...

Found ...

PHKL_33457 s700_800 11.11 asyncio driver patch

For some additional information ...

1. The asyncio driver now supports a new device file minor
number 256. If an application uses the async device
file with this new minor number, the driver will defer
locking the memory until I/O's are issued. This avoids
the overhead of memory setup time and thus reduces the
application startup time.

Note: This minor number should only be used on systems
that have enough physical memory so that paging is
avoided. Paging can cause serious performance degradation
with this new enhancement. On systems where paging is an
issue, this minor number should not be used.

Command-Line Junkie
Steven M Evans
Occasional Advisor

Re: Sybase 12.5, HP-UX 11.11 Kernel ... Tuning ...

Here's some info from a ticket I sent to EMC:

After reading emc26766 and emc58837, I would like a confirmation of my understanding of timeout values for LVM and minor numbers for the /dev/async device.
First, our LVM timings are a 90 second timeout for pvs and 0 for lvs for all connections to Symmetrix through powerpath. This corresponds with emc58837. These settings are in place for Sybase as well. emc26766 states that timeouts for Sybase should be:
1. lvchange -t 180 /dev/VGXX/lvXXX of all lvols.
2. pvchange -t 30 of all physical volumes. Is this correct for our systems?

Second, we have a non-mirrored Sybase database; therefore, I believe our minor number for the async device should be 0x0 or 0x2. Since the recommendation from emc26766 turns off immediate reporting for a mirrored database, should this also be followed for a non-mirrored database? If this is the case, then the minor number should be 0x0. However, the setting of 0x2 is specific for Sybase and may shorten system calls. How does this setting effect the Symmetrix?

EMC recommends that for hosts running power path that you set the pvtimeout to 90 and not control timeouts at the lv level. So the lvtimeout would be 0 << or since this is the default, don't set the lvtimout.

As to setting async - since you are not seeing the error discribed in emc26766 I would disregard that primus. Its my understanding that you are going to be talkin to Sybase re: async. My suggestion is to take Sybase's advice and set async to whatever they recommend.

I discussed settings for the async driver with EMC and the impact it could have on the drives. Here├в s a summary of each minor number bit and their recommendation:

0x1 ├в Since this is currently not set, do not change it. I/O errors would show up in the Sybase log if this needed to be changed.

0x2 ├в EMC was aware of this setting and agreed it could speed up I/O. They actually warned against using it with the older disk, since the they would not be able to handle it.

0x4 ├в Since Sybase mirroring is not currently used, do not change. Changing this may result in data corruption.

0x256 ├в Does not effect EMC. I found this setting on the forums and in patch PHKL_33457. This will defer locking until I/O├в s are issued and should speed up startup.



This is this.