Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

Yes, I use Oracle DB mainly for OLTP during the day but it has some OLAP during night for reporting. I found some disk has high read busy. I asked programmer and they said they do read data from one table and insert into another table. So they want to split disk into 2 LUN, one LUN for read table and another LUN for write table but I think it should not help too much because whole LUN will be striped into all back-end SSD in SAN storage I do not expect I will get double performance from 2 LUN. Because I do not think split 1 large LUN into 2 smaller LUN will help much in performance improvement. What do you recommend? 

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

Khun Kritsada

For OLTP and OLAP workload, I would say having more LUNs is preferred.  For the OS, each disk unit (LUN in this case) provides one disk IO queue.  Having more queues for disk IO is better than having less when you know there are a lot of disk IOs at its peak workload.

Does the SAN disk server provide SSD/Flash drives or spinning disks? 

When you you saw high disk % Busy, what was the high value range? Was it lower or higher than 50% busy? 

If you use all SSD/Flash Disk from SAN disk server and see disk % Busy consistently higher than 20%, I would say there is some serious performance bottleneck in your SAN server. Does the SAN server use Data Reduction Pool which imposes performance overhead that can cause disk response time degradation at high disk write workload?    For all spinning disks, disk % Busy higher than 50% can be considered bad.

Other factors can affect SAN disk performance ans you can study these from the following articles: 

Best Practices for AIX in a SAN Environment: Part 1 at  https://techchannel.com/IT-Strategy/06/2021/best-practices-aix-san

Best Practices for AIX in a SAN Environment: Part 2, I/O Configurations at  https://techchannel.com/SMB/07/2021/san-best-practices-i-o-configurations

Yes, I use Oracle DB mainly for OLTP during the day but it has some OLAP during night for reporting. I found some disk has high read busy. I asked programmer and they said they do read data from one table and insert into another table. So they want to split disk into 2 LUN, one LUN for read table and another LUN for write table but I think it should not help too much because whole LUN will be striped into all back-end SSD in SAN storage I do not expect I will get double performance from 2 LUN. Because I do not think split 1 large LUN into 2 smaller LUN will help much in performance improvement. What do you recommend? 

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

HI K.Satid,

Thanks for your response. Our SAN storage use SSD as a back-end. I have just single 100GB OS LUN and I have not a problem with OS LUN. I just want to make sure about meaning of high disk busy in topas because I think it just tell me this high disk busy has more workload compare to another disk. If I want to know this high disk busy has very busy I must see it at SAN storage level, not OS level because new modern SAN storage will do stripe LUN to across all their back-end SSD in SAN storage. Am I right regarding about meaning of TOPAS disk busy in OS point of view? 

My problem is programmer do SQL statement to read data in one table and write into another table and I saw high disk busy and high read operation and both tables are in the same LUN.  Actually, I try to study Oracle AWR report to investigate SQL statement and with my old knowledge is if it has problem for some query statement, not whole database instance then AWR cannot help but need to see execution plan instead. So my programmer want to split one large LUN to 2 smaller LUN, one LUN for keep read table and another LUN for write table. Just easy explain what programmer want to do.

I do not think split into 2 smaller LUN will help much because LUN is already striped to all back-end SSD. One thing that I am not sure in my original posted is queue depth, if I have more LUN then I have more queue depth, right? Regarding single large LUN, If we are not run out of LUN queue depth then spit into 2 smaller LUN will not help much IO performance improvement, right? Any things I miss to consider regarding about one large LUN or more smaller LUN?

------------------------------
Kritsada Ruayruengrung

Original Message:
Sent: Wed November 01, 2023 11:03 PM
From: Satid Singkorapoom
Subject: DISK Busy means in new modern SAN storage via TOPAS

Khun Kritsada

For OLTP and OLAP workload, I would say having more LUNs is preferred.  For the OS, each disk unit (LUN in this case) provides one disk IO queue.  Having more queues for disk IO is better than having less when you know there are a lot of disk IOs at its peak workload.

Does the SAN disk server provide SSD/Flash drives or spinning disks? 

When you you saw high disk % Busy, what was the high value range? Was it lower or higher than 50% busy? 

If you use all SSD/Flash Disk from SAN disk server and see disk % Busy consistently higher than 20%, I would say there is some serious performance bottleneck in your SAN server. Does the SAN server use Data Reduction Pool which imposes performance overhead that can cause disk response time degradation at high disk write workload?    For all spinning disks, disk % Busy higher than 50% can be considered bad.

Other factors can affect SAN disk performance ans you can study these from the following articles: 

Best Practices for AIX in a SAN Environment: Part 1 at  https://techchannel.com/IT-Strategy/06/2021/best-practices-aix-san

Best Practices for AIX in a SAN Environment: Part 2, I/O Configurations at  https://techchannel.com/SMB/07/2021/san-best-practices-i-o-configurations

------------------------------
Chance favors only the prepared mind.
-- Louis Pasteur
------------------------------
Satid S.

Original Message:
Sent: Wed November 01, 2023 10:05 PM
From: Kritsada Ruayruengrung
Subject: DISK Busy means in new modern SAN storage via TOPAS

Yes, I use Oracle DB mainly for OLTP during the day but it has some OLAP during night for reporting. I found some disk has high read busy. I asked programmer and they said they do read data from one table and insert into another table. So they want to split disk into 2 LUN, one LUN for read table and another LUN for write table but I think it should not help too much because whole LUN will be striped into all back-end SSD in SAN storage I do not expect I will get double performance from 2 LUN. Because I do not think split 1 large LUN into 2 smaller LUN will help much in performance improvement. What do you recommend? 

------------------------------
Kritsada Ruayruengrung

Original Message:
Sent: Wed November 01, 2023 03:35 PM
From: Tom McGivern
Subject: DISK Busy means in new modern SAN storage via TOPAS

Yes, the "disk" you see in tapas (if communicating to the SAN), is just another virtualization layer. But, it's what AIX knows about, so that's all it can report.

as far as sizing and quantity, it really depends on your target I/O characteristics.
Are you running an OLTP database? or a deep-freeze archive storage space?


Original Message:
Sent: 11/1/2023 9:02:00 AM
From: Kritsada Ruayruengrung
Subject: DISK Busy means in new modern SAN storage via TOPAS

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

------------------------------
Kritsada Ruayruengrung
------------------------------

Yes, I use Oracle DB mainly for OLTP during the day but it has some OLAP during night for reporting. I found some disk has high read busy. I asked programmer and they said they do read data from one table and insert into another table. So they want to split disk into 2 LUN, one LUN for read table and another LUN for write table but I think it should not help too much because whole LUN will be striped into all back-end SSD in SAN storage I do not expect I will get double performance from 2 LUN. Because I do not think split 1 large LUN into 2 smaller LUN will help much in performance improvement. What do you recommend? 

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

Yes, I use Oracle DB mainly for OLTP during the day but it has some OLAP during night for reporting. I found some disk has high read busy. I asked programmer and they said they do read data from one table and insert into another table. So they want to split disk into 2 LUN, one LUN for read table and another LUN for write table but I think it should not help too much because whole LUN will be striped into all back-end SSD in SAN storage I do not expect I will get double performance from 2 LUN. Because I do not think split 1 large LUN into 2 smaller LUN will help much in performance improvement. What do you recommend? 

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks 

I agree with Satid that you want to spread IO over several hdisk for a file system or ASM DG to be able to distribute IO over all resources in your SAN, not just within the physical storage sub-system, as well as not being constrained by the hdisk queue depth in AIX.

For example, for an OLTP database I typically spread IO over at least 8 LUNs / hdisks for a production DB and utilize maximum queue depth (potentially constrained by storage subsystem limitations!). 

You may want to check my session on AIX VUG from August, 27th 2020 you can find here: https://www.ibm.com/support/pages/power-systems-virtual-user-group - scroll down to past sessions. That session Oracle Best Practices Part 2 (I/O & Miscellaneous) discusses this topic in more detail.

Regarding new modern SAN storage, it does striping pieces of LUN to the whole back-end SSD. If I use TOPAS and I see higher disk busy% so I think it just means this disk just has more workload than another disk but it does not means very busy because new modern SAN storage will stripe LUN to the whole back-end storage and if I want to know it has real busy then I think I must see at SAN storage perspective point of view. Am I right? If not please help me interpret the meaning of disk busy. Another questions related to new modern SAN storage. Should I have one large LUN or many smaller LUN? Are there any things should I use to consider regarding the number of LUN? Sample queue depth, database or application, filesystem type?

Thanks