UNSOLVED
Symstat IO/sec vs Requests/sec
Hi guys,
I posted last week with a symstat/WP query that was sorted out in short order so now I'm back with another
What I'm doing is working my way through chapter 2 of a doc called 'EMC ControlCenter Performance Manager : The Symmetrix In Pictures' which is a cracking good document & I have a question around monitoring IO's vs Requests.
Just to be difficult I don't currently have access to ECC in my environment so what I'm doing is trying to translate the metrics discussed in the above document into SYMCLI/symstat commands that report the relevant info.
This question is around data alignment/IO's vs Requests.
The document says that the number of IO/sec coming from an FA should match pretty closely the number of Cache Requests/sec, hence indicating that the IO's are aligned on the cache slot boundaries.
Here's an edited symstat from our system showing 157 IO/sec for FA8B the number of cache requests is 10 read + 166 write.
The second line shows 23 IO/sec for FA8C with 1 read + 23 write cache requests/sec.
# symstat -sid 252 -type requests -dir all -i 5
DIRECTOR IO/sec Cache Requests/sec % RW
17:05:11 READ WRITE RW Hits
FA-8B 157 10 166 176 100
FA-8C 23 1 23 24 100
Can I assume from this very small subset of samples that the IO's are pretty much aligned & therefore not a problem ?
Now, given that my Sym is performing fine, I'm not worried about that currently, but want to make sure I know what unaligned IO looks like if I ever do see it so I've tried to mock up the output below........
Am I on the right sort of track here ? The document mentions that unaligned IO may use more than one cache slot & force a read to calculate CRC.
IE 157 IO/sec for FA8B would have reserved 416 cache slots which is way more than one per IO. The document notes that if you see requests reported as double the IO count then this can indicate unaligned IO.
# symstat -sid 252 -type requests -dir all -i 5
DIRECTOR IO/sec Cache Requests/sec % RW
17:05:11 READ WRITE RW Hits
FA-8B 157 250 166 416 100
FA-8C 23 23 23 46 100
If you guys could double check my logic &/or suggest any better symstat command for zeroing in on this type IO/sec vs Cache Requests/Sec info I'd really appreciate it.
Of course if anyone wants to ignore the question & poke fun at me for thinking about this kind of geeky stuff on a perfectly good Saturday when I should be doing normal things then that's fine too
I have done some normal stuff too but seem to have slipped into Geek mode over the last hour :-p
Many thanks
I posted last week with a symstat/WP query that was sorted out in short order so now I'm back with another
What I'm doing is working my way through chapter 2 of a doc called 'EMC ControlCenter Performance Manager : The Symmetrix In Pictures' which is a cracking good document & I have a question around monitoring IO's vs Requests.
Just to be difficult I don't currently have access to ECC in my environment so what I'm doing is trying to translate the metrics discussed in the above document into SYMCLI/symstat commands that report the relevant info.
This question is around data alignment/IO's vs Requests.
The document says that the number of IO/sec coming from an FA should match pretty closely the number of Cache Requests/sec, hence indicating that the IO's are aligned on the cache slot boundaries.
Here's an edited symstat from our system showing 157 IO/sec for FA8B the number of cache requests is 10 read + 166 write.
The second line shows 23 IO/sec for FA8C with 1 read + 23 write cache requests/sec.
# symstat -sid 252 -type requests -dir all -i 5
DIRECTOR IO/sec Cache Requests/sec % RW
17:05:11 READ WRITE RW Hits
FA-8B 157 10 166 176 100
FA-8C 23 1 23 24 100
Can I assume from this very small subset of samples that the IO's are pretty much aligned & therefore not a problem ?
Now, given that my Sym is performing fine, I'm not worried about that currently, but want to make sure I know what unaligned IO looks like if I ever do see it so I've tried to mock up the output below........
Am I on the right sort of track here ? The document mentions that unaligned IO may use more than one cache slot & force a read to calculate CRC.
IE 157 IO/sec for FA8B would have reserved 416 cache slots which is way more than one per IO. The document notes that if you see requests reported as double the IO count then this can indicate unaligned IO.
# symstat -sid 252 -type requests -dir all -i 5
DIRECTOR IO/sec Cache Requests/sec % RW
17:05:11 READ WRITE RW Hits
FA-8B 157 250 166 416 100
FA-8C 23 23 23 46 100
If you guys could double check my logic &/or suggest any better symstat command for zeroing in on this type IO/sec vs Cache Requests/Sec info I'd really appreciate it.
Of course if anyone wants to ignore the question & poke fun at me for thinking about this kind of geeky stuff on a perfectly good Saturday when I should be doing normal things then that's fine too
Many thanks
404
0
Posted February 23rd, 2008 13:00
Hey there RRR, someone else who's thinkin' about this stuff at the weekend. Hurray.
As I said, the solutions guide document/Symm in pictures that I'm currently reading is one of those great documents that don't come along that often about the Sym. It's really helping me to think about how some of the components work together in more detail than I have up until this point.404
0
Posted February 23rd, 2008 14:00
Is that pictures document publically available ? Or maybe for partners somewhere ?404
0
Posted February 23rd, 2008 15:00
I'm not sure, & now you've asked I can't currently find a link to it on Powerlink.
Unless anyone else can shed any light I'll try to check tomorrow.404
0
Posted February 24th, 2008 03:00
Guys it looks like you all have a lot of spare time .. Why don'y you fly here ... I have to clean my house ..
I hope Rob won't beg for points .. at least not 'till I can try to answer the question seriously
Cheers .. and see you tomorrow ..404
0
Posted February 24th, 2008 08:00
I promise.... I think....404
0
Posted February 24th, 2008 08:00
I searched as well, but cannot find it publically or in the partner section404
0
Posted July 26th, 2008 05:00
Hi,
To compare the request/s and IO/s on a FA, you have to check three things.
1- The average IO size coming to the FA : If the average IO size(block size) is bigger than a cache slot(32K for older DMX, 64K for DMX3/4), for example 128K, then you have double number of the requests/s compared to the IO/s.
2- Misalignment: Your case. If the average IO size is 64K on a DMX3/4 and your requests/s is higher than IO/s then you probably have misalignment. But as I know that is only possible on Windows and Solaris hosts.
3- Locality of Reference: Look at this subject on that book. If you average IO size is smaller than a cache slot or if your application uses locality of reference, then you will probably have more IO/s compared to the requests/s.
In your case, nothing can be said without checking the IO size. If you have Performance Manager, you can check for a long time period. If you do not have, you can have an average size with symstat also for a short time period.404
0
Posted August 1st, 2008 02:00
v760: is your question answered or do you still have questions ?
If you're fine, please mark the question as answered and reward the posts with points for being "correct" and "helpful".232
0
Posted August 1st, 2008 03:00
Rob I supposed you w2s still on vacation .. Do you miss our forums ??
Message was edited by:
Stefano Del Corno
RRR
6 Operator
•
5739 Posts
404
0
Posted February 23rd, 2008 11:00
Uhm.... I'm very interested in performance stuff myself, but I regret that I didn't come up with this question myself....
I wonder what this thread is going to look like and certainly hope some performance gurus will join us here with some good answers !!!
C'mon guys: let the games ... uhm.... fun begin !