Re: New HDB backend MDB

Howard Chu wrote:
> Nico Williams wrote:

>> What makes your write throughput so slow relative to BDB?

One more comment - my initial write throughput tests, as reported in the 
LDAPCon paper, failed to take transaction deadlocks into account. I.e., in the 
data set I used, there were zero deadlocks using BDB or MDB. But in a 
realistic LDAP installation with various indexing configured, index updates 
with BDB run into deadlocks all the time. Retested under these conditions, 
MDB's throughput is higher than BDB's because it never deadlocks and thus 
never needs any deadlock recovery/retry steps.

>>  Do you
>> fsync() more often than you have to?  When do you fsync()?

> On a per-write basis we are faster than BDB, because we do less work. No
> buffer copying...

-- 
   -- Howard Chu
   CTO, Symas Corp.           http://www.symas.com
   Director, Highland Sun     http://highlandsun.com/hyc/
   Chief Architect, OpenLDAP  http://www.openldap.org/project/