Unsolved
This post is more than 5 years old
138 Posts
0
1041
2 Gb/s paths versus 4 Gb/s paths
All;
I want to ask your opinion.
We have a HP-UX Server with Oracle DB with 4 x HBAs. 2 are 2 Gb/s and 2 are 4 Gb/s. So, naturally, we only used the 2 x 4 Gb/s HBAs - they are faster.
Our Oracle DBAs had a performance problem and talked to the UNIX Sys Admins, who said that the Server was queuing I/O up at the paths to the disk volumes. They wanted more paths from the Servers to the disks.
I looked at the VMAX, and it DOES NOT have any performance problems. All looks good.
So, should I:
1. add 2 more paths to the existing 4 Gb/s HBAs, i.e.- zone and create masking views to 2 x new FA ports on the SAME 4 Gb/s HBAs, or
2. should I use my 2 x 2 Gb/s HBAs?
I think that with 1) I would have conflict at the HBAs, but the data would go "faster". With 2) I wouldn't have any contention on the HBAs, but the data would go "slower".
Are there any EMC best practices that say that the same device should only be accessed at the same speed, or anything like that?
(Please don't think that this is trivial, we have a SERIOUS disagreemnt here about this.)
Stuart
Quincy561
1.3K Posts
1
September 1st, 2011 09:00
For small block IOs. 4gb vs. 2gb really makes no difference.
If you are concerned about the imbalance, add the 2 extra HBAs and set the speed of the 4gb HBAs to 2gb.
How many FAs are the HBAs connnected to? 1 to 1 mapping?
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
1
September 1st, 2011 09:00
HBAs themselves are not overloaded right ? If you look at port utilization on the switch side ?
StuartA58
138 Posts
0
September 1st, 2011 10:00
There were 1:1. Now I have 2 x HBAs pointing at same FAs. It's like this now:
HBA1 (2 Gb/s)
HBA2 (2 Gb/s)
HBA3 (4 Gb/s) - 12g1
HBA4 (4 Gb/s) - 5g1
HBA5 (4 Gb/s) - 10f1 12g1
HBA6 (4 Gb/s) - 7f1 5g1
5449Pr4QlS12567
46 Posts
1
September 1st, 2011 11:00
I too would recommend to check port utilization. I remember reading somewhere on powerlink to keep the same model/firmware/speed HBAs on a host.
I would map the devs to an extra set of FAs and zone them to existing 4Gb Hbas. This will increase the no. of Queues available for I/O. This might not certainly solve the issue but you can give it a try.
5449Pr4QlS12567
46 Posts
0
September 1st, 2011 11:00
Then you also might wan to check the internal disks if host is using any.
Message was edited by: Deepak Malhotra
StuartA58
138 Posts
0
September 1st, 2011 11:00
HBAs are NOT overloaded. % Util is:
2 Gb/s HBA 0%
2 Gb/s HBA 0%
4 Gb/s HBA 20%
4 Gb/s HBA 10%
5449Pr4QlS12567
46 Posts
0
September 1st, 2011 11:00
That was the suggestion b4 I saw the HBA % util spec
Quincy561
1.3K Posts
0
September 1st, 2011 11:00
If the HBAs are not overloaded, why add more?
StuartA58
138 Posts
0
September 2nd, 2011 13:00
Thanks for help.
It turns out that we do have some servers with 2 Gb/s and 4 Gb/s paths to the same VMAX devices. It seems to work, and we don't have any problems.
We added paths to this server, but Month-End-Closing is over on this server, so usage went way down. We won't know if we have this perofrmance problem again until next month.
Stuart
StuartA58
138 Posts
0
September 2nd, 2011 13:00
Apparently it's a HP-UX thing. HP-UX maintains a queue for each "path". More paths (to the sam disks) means more concurrent I/O.
dynamox
1 Rookie
1 Rookie
•
20.4K Posts
0
September 2nd, 2011 17:00
have you experimented with increasing queue_depth parameter ?
scsictl -m queue_depth= /dev/rdsk/
I think default is 8, i would try to double it.
StuartA58
138 Posts
0
September 3rd, 2011 08:00
Thats a good idea. I'll talk to the unix boys on Monday!!!!
Sent from my iPhone
AJK3
108 Posts
0
September 3rd, 2011 10:00
as dynamox mentioned Q depth should help out if there are seeing queuing of IO