Wim Coekaerts
2004-Jun-21 13:21 UTC
[Ocfs-users] performance when using securepath or powerpath
For those customers that are on this list and are using EMC Powerpath or HP Securepath and have performance problems while using ocfs and point to ocfs first. the actual io overhead for an io on ocfs compared to raw, when it's database io is negligable, we don't do any locking, we have about 40000 clockticks of processing from logical to physical offset, nothingin todays cpu world. Every real setup with a stable io stack has shown that we are at the same performance as raw io, we use up about 1.5% more cpu but that's it. In many cases where customers had complaints about performance it turned out to be the use of powerpath or securepath in the system stack that caused the problem. Some of these things have been fixed by those vendors, however we canot help as it's not open source. If you have performance issues, and you use those products, at the very least, run the same tests WITHOUT them loaded. and go from there. it will save you a lot of time and fingerpointing. One other potential problem is using badly compiled device drivers, if an FCdriver does not use HIGHMEMIO or Varyio, then the performance will be very bad, so another suggestion would be to at least give the distribution shipped (rhat or suse) fc driver a chance as well. even if it's not certified or supported at least give it a try to compare. this is just a word of advice. We get too many such questions that always end up being down below in the stack. it's not to say those 3rd party products are -bad-, but it's a black box so take it out first and go from there.