Re: Writing at 847 MB/sec ???
Posted by:
Rick James
Date: April 19, 2015 01:38PM
Even after adjusting for "marketing inflation", the numbers seem "wrong". Suggest you approach FusionIO with your poor numbers.
What is the value of innodb_flush_log_at_trx_commit? The value of 1 is safer, but slower. (This applies on writes to the database, so only on "Restore".)
What about the "double-write" buffer? As I understand it, it can be turned off if the FusionIO card is set up with atomic 16KB writes and the file system is aligned appropriately. (Sorry, I don't know the config details.) This eliminates a significant number of writes for FusionIO that should be kept on HDD.
Clarification... You said "Before using the Fusion ioDrive as a cache". Are you using Fusion as a _cache_ in front of the HDD? Or is it the mass media?
As a cache, things like "write-through" versus "write-back" need to be specified. Then we need to discuss "caching". Backup & Restore may be the worst choice for a Benchmark because almost nothing can be "cached".
If Fusion is being used as the only mass storage, then I would expect 10x easily and the marketing quote of 40x is 'not impossible'.
Subject
Views
Written By
Posted
1419
April 11, 2015 05:50AM
804
April 15, 2015 11:03PM
787
April 17, 2015 08:45AM
Re: Writing at 847 MB/sec ???
709
April 19, 2015 01:38PM
764
April 20, 2015 01:00AM
Sorry, you can't reply to this topic. It has been closed.
Content reproduced on this site is the property of the respective copyright holders.
It is not reviewed in advance by Oracle and does not necessarily represent the opinion
of Oracle or any other party.