Does allocation performance vary by collector?

[Matt Khan]

>> And I totally agree with Ramki that just showing averages might not be 
very productive and you'll be better off showing distribution statistics.
I do have distributions (I output a centile distribution basically) but am 
quoting averages because 

a) I'm not sure what an easy way to distribute this information is over a 
mailing list, and 
b) there is a consistent pattern around the quoted mean

with respect to b), the distribution for each looks very much the same 
except the average is shifted... it's reasonably tight around the mean 
given the hardware we run on (typically +/- 100 micros for 90% then a bit 
of a tail but not tragic if no STW pauses).

>> To increase the heap region size you can use the G1HeapRegionSize=N 
parameter. Currently, the allowed range is between 1m and 32m.
is there a guide to G1 tuning anywhere? I haven't come across one hence 
why I've used default values (and perhaps why we appear unable to get it 
as fast as cms).

>> modulo the caveats about NUMA-allocator anti-patterns above.
I intend to change our main case that shows no gain with a NUMA-allocator 
so that the longer lived object is created in the thread that uses it so 
will repeat the bench at that point.

Cheers
Matt

Matt Khan
--------------------------------------------------
GFFX Auto Trading
Deutsche Bank, London


---

This e-mail may contain confidential and/or privileged information. If you are not the intended recipient (or have received this e-mail in error) please notify the sender immediately and delete this e-mail. Any unauthorized copying, disclosure or distribution of the material in this e-mail is strictly forbidden.

Please refer to http://www.db.com/en/content/eu_disclosures.htm for additional EU corporate and regulatory disclosures.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20100414/59392d0d/attachment.htm>