I'm building some new SAS servers on a Power8.
And I'd like to get i right, for best performance.
The Compute server, has a /saswork mountpoint, for temporary data.
I have followed the tuning guide, and then they mention this:
For SAS WORK area file systems which store only temporary data, consider turning off logging to improve performance.
# mount -o log=NULL /mountpoint
Does any of you, have any expirience with this?
Does it make a huge difference in performance?
Other sugestions?
My SAS Compute server - so far (SAS not installed):
OSlevel 7200-01-01-1642
7 cores - SMT 8
56 GB RAM
Each filesystem, has it's own Volume Group, with seperate disk for JFS2LOG:
Filesystem GB blocks Free %Used Iused %Iused Mounted on
/dev/saslv 249.88 249.84 1% 4 1% /sas
/dev/sasworklv 249.88 249.84 1% 4 1% /saswork
/dev/sasutillv 199.88 199.84 1% 4 1% /sasutil
/dev/sasdatalv 1497.50 1497.27 1% 4 1% /sasdata
/dev/spdsdatalv 1048.25 1048.09 1% 4 1% /spdsdata
mounted mounted over vfs date options
--------------- --------------- ------ ------------ ---------------
/dev/saslv /sas jfs2 Mar 03 08:33 rw,rbw,rbr,noatime,log=/dev/sas_loglv
/dev/sasworklv /saswork jfs2 Mar 03 08:33 rw,rbw,rbr,noatime,log=/dev/saswork_loglv
/dev/sasutillv /sasutil jfs2 Mar 03 08:33 rw,rbw,rbr,noatime,log=/dev/sasutil_loglv
/dev/sasdatalv /sasdata jfs2 Mar 03 08:33 rw,rbw,rbr,noatime,log=/dev/sasdata_loglv
/dev/spdsdatalv /spdsdata jfs2 Mar 03 08:33 rw,rbw,rbr,noatime,log=/dev/spdsdata_loglv
SAS has it's own HBA's.
These tuning settings are also implemented:
ioo -p -o j2_dynamicBufferPreallocation=256 -o j2_maxPageReadAhead=2048 -o j2_minPageReadAhead=16 -o j2_nPagesPerWriteBehindCluster=64
vmo -o nokilluid=10
no -o tcp_nodelayack=1
chdev -l sys0 -a maxuproc=2048
lvmo -v rootvg -o pv_pbuf_count=2048
apply "lvmo -v %1 -o pv_pbuf_count=1024" $(lsvg | grep -v root)
maxfree = 15176
minfree = 960
Thankyou in advance 🙂
Chears
Harald
Simply test it. Logging in journaling file systems is designed to prevent data loss in case of a system crash / unexpected hard shutdown / power loss. Since SAS work files would be lost anyway by "definition" in such a case, your only penalty might be a longer fsck when the system recovers, or a possible rebuild of the WORK file system from scratch if the damage is too much.
So create a real burner of a SAS program that makes heavy use of the WORK, run it once with standard settings, activate the mount -o log=NULL /mountpoint, restart the system and rerun the program (both runs with fullstimer).
Simply test it. Logging in journaling file systems is designed to prevent data loss in case of a system crash / unexpected hard shutdown / power loss. Since SAS work files would be lost anyway by "definition" in such a case, your only penalty might be a longer fsck when the system recovers, or a possible rebuild of the WORK file system from scratch if the damage is too much.
So create a real burner of a SAS program that makes heavy use of the WORK, run it once with standard settings, activate the mount -o log=NULL /mountpoint, restart the system and rerun the program (both runs with fullstimer).
Hi Kurt,
Thank you for your response.
OK - I'll wait for SAS to install their software, and I will run the tests with them.
If it should crash and it's only temprary data, maybe it's faster to remove the filesystem, and recreate it, just with subfolders.
Best Regards
Harald
It has been my experience that AIX (our DWH runs on AIX since 2000, started with AIX 4.3.2) is very reliable and hardened against outages.
We only lost a filesystem once, and that was due to a not-very-well-tested, bleeding-edge storage virtualization software (not from IBM) that scrambled our mirrors.
But I have to admit that I keep jfs2 logging on, as I have devised a multi-disk work storage layout that allows users to create semi-permanent data that persists from SAS session to SAS session, but has a limited "shelf life". So I want those filesystems to survive an outage without data loss.
I see your point, and I will disguss this with our SAS users.
Though, I will still want to make the test, just to see the difference.
Then it's up to the SAS users, what to choose.
Thanks
Harald
Hello Margaret,
I'm not in a position to have any expences with this matter.
But if one of your AIX experts, free of charge, could look through my post, I would be happy.
We are building a 14 core SAS system on Power 8 and AIX 7.2. And I'd like to have MAX performance 🙂
Cheers,
Harald
Completely forgot what @MargaretC mentioned: the tuning guides and whitepapers that SAS publishes with help from the system manufacturers are first-class, and one can see throughout that the suggested settings are the result of thorough testing.
I gleaned a lot of useful information from them when setting up our server. Without that, I wouldn't be able to run it with considerably aged HW as it is now. Will of course consult the lastest papers when setting up the next incarnation.
Hello Margaret,
I have dual VIO servers, only for SAS.
Each has a 16Gbs HBA, but it can only run 8Gbs, due to Storage capability.
Storage is SAN devices, connected through SVC:
hdisk100 Available C7-T1-01 MPIO FC 2145
devices.sddpcm.71.rte 2.6.6.0 COMMITTED IBM SDD PCM for AIX V71
Can't go much higher, due to another old system connected to SVC.
The LUNs are ordered as Tier 1 on an IBM V7000 SAN storage.
Regarding SAS WORK in memory - No, we won't do that.
Cheers
Harald
Here is what my AIX expert said about your storage:
"well, I'm fine with him using SVC, under the right conditions. We generally don't consider V7000 as Tier1 disk either, as it's our mid-range storage offering.
I'm not seeing any disk striping here. He's only showing 1 hdisk on the host side, and No, that is not good. We need to surface multiple LUNs for EACH filesystem at the AIX level. It's never advised to even use only 1 hdisk. Also, with 7 cores in Compute, we're going to say we need to drive around 1 GBps of throughput. A single 16Gb HBA running at 8 Gb will never get them there. We tend to use 750MBps as the clip point.
I'm not sure why they're using SDD-PCM for multi-pathing though. Generally, AIX MPIO is used, it's been the default for AIX for many years. . Maybe SVC is driving them there, but I've done SVC using MPIO, I believe. "
Please review this BP for IO Configure for SAS paper so that you will understand how SAS does IO and what needs to be in place to get MAX performance. http://support.sas.com/resources/papers/proceedings16/SAS6761-2016.pdf
I think we need to take this off the public chat and have a meeting with my IBM expert and yourself so we can discuss this via a conference call.
Cheers,
Margaret
Arghhh - Sorry, my fault.
No, I'm using several LUNs for my filesystems.
mkvg -S -s 128 -y sas_vg hdisk100 hdisk101
mkvg -S -s 128 -y saswork_vg hdisk200 hdisk201
mkvg -S -s 128 -y sasutil_vg hdisk300 hdisk301
mkvg -S -s 256 -y sasdata_vg hdisk400 hdisk401 hdisk402 hdisk403 hdisk404 hdisk405 hdisk406 hdisk407 hdisk408 hdisk409 hdisk410
mkvg -S -s 256 -y spdsdata_vg hdisk500 hdisk501 hdisk502 hdisk503 hdisk504 hdisk505 hdisk506 hdisk507
JFS2LOG is placed on hdisk[1-5]00
Filesystems on *01 disks and up.
We use SDDPCM to use all paths on the adapter.
A conference call would be fine, thanks.
Cheers
Harald
@HejHarald wrote:
PS. Sorry for my outburst (the Arghh). It was only ment for myself 🙂 ( I pressed Post, before thinking )
Don't bother. After all, this is a IT community, and we all have those moments where we just need to vent before exploding 😉
The SAS Users Group for Administrators (SUGA) is open to all SAS administrators and architects who install, update, manage or maintain a SAS deployment.
SAS technical trainer Erin Winters shows you how to explore assets, create new data discovery agents, schedule data discovery agents, and much more.
Find more tutorials on the SAS Users YouTube channel.