Re: Question on IO utilization capacity

Basically in one of the system(X5 machine) with write back flash cache , we are seeing the flash cache IO is causing bottleneck , mainly large reads are flooding the flash cache. Whereas in the other system(X3 machine ) with write through flash cache mode, we are seeing hard disk IO is reaching its max limit(mainly small writes are coming on top). I understand its blunt one, but has anybody encountered such symptoms or just by looking into such symptoms , is this pointing towards capacity issues which could be addressed by increasing either the number of storage cells or by increasing flash cache size for each storage cell?

On Wed, Jan 6, 2021 at 10:18 PM Lok P <loknath.73_at_gmail.com> wrote:

> Hi All, we are on Oracle Exadata image version 19.2.9.0 but the database
> is still on 11.2.0.4. It's a X5-2 , half RAC machine with ~7 cell servers
> and ~41TB of flash cache(i.e. ~5.9TB of flash cache per cell disk, write
> back). Its only one database resides in this exadata machine.
>
> Just to give a background, many times we encounter situations where the
> flash cache response time increases to ~10ms for the small reads and impact
> the oltp workload, and it happens mainly when we have a lot of full
> scan/smart scans happen from the warehousing type workload at same time.
> And during this the flash disk Io utilization reaches up to ~80% with large
> reads showing ~30K IOPS. Similarly another exadata box on X3-2 with FULL
> RAC with multiple databases mounted on it though and ~20TB flash
> cache(Write through), and we are seeing a spike in hard disk response/IO
> utilization but not such spike in flash disk response or IO utilization.
>
> Now we are definitely working on making the application queries better but
> as part of tech refresh, the infra team already has a plan to move to a new
> exadata box , so management is asking, if having a full rack or additional
> storage cell servers added, would be helpful in making the above situation
> better? Or say having more flash cache allocated with the same number of
> cell servers will help us create some more capacity? Want to understand
> experts' thoughts on the same.
>
>
> Regards
>
> Lok
>

--
http://www.freelists.org/webpage/oracle-l