[PATCH] LMDB full patch set
metze at samba.org
Fri May 18 11:13:03 UTC 2018
Am 18.05.2018 um 10:34 schrieb Andrew Bartlett via samba-technical:
> On Thu, 2018-05-17 at 16:15 +1200, Garming Sam wrote:
>> I suspect Joe's concurrent stress-testing would tell us more about how
>> it behaves under real load -- we definitely should block less now,
>> although not as little as LMDB fully allows due to other requirements.
> Just to follow up on this with some numbers from an internal thread.
> We got a load of 167 (I think operations / second) on lmdb, essentially
> the same as the 160 we saw on master. Windows 2012R2 gives 235 on the
> same virtual hardware, for reference.
> So no massive gain, but no massive loss either. This will likely
> change with larger user sets (I don't have the details to hand as to
> what the DB size was).
Thanks! I also compared autobuild times and got similar results.
One thing I was wondering about is if we should make use of virtual
databases within lmdb. We could separate the main records from the
index records. In theory we should gain a lot for unindexed search
as we can traverse only the real records. I guess the relationship from
records to corresponding index records (every user has (at least) more
than 5 index records) gets more impact with larger databases.
For 100000 objects in the database we would at least have 500000
index records and skipping them efficiently during an non-indexed search
is a lot!
I'm not asking to do that before we can push this, but we should
at least have ways for a seamless change in future. The compatibility
flags at the samba_dsdb level are not a valid way to detect this!
We need this at the ldb_lmdb layer.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 836 bytes
Desc: OpenPGP digital signature
More information about the samba-technical