[ https://issues.jboss.org/browse/JGRP-2017?page=com.atlassian.jira.plugin.... ] Bela Ban edited comment on JGRP-2017 at 2/22/16 1:14 PM: --------------------------------------------------------- OK, I'll take a look whether byte[] arrays can be reused, too. Something to try out might be to replace the payload of the existing messages *in-place* and add a {{CompressHeader}} if successful. We have to consider the following cases: h5. {{COMPRESS}} is _above_ {{NAKACK2}} / {{UNICAST3}} The message is added to {{NAKACK2}} / {{UNICAST3}} with a compressed payload and a {{CompressHeader}} attached. {{NAKACK2}} / {{UNICAST3}} will retransmit the _compressed_ message h5. {{COMPRESS}} is _below_ {{NAKACK2}} / {{UNICAST3}} {{NAKACK2}} / {{UNICAST3}} will retransmit the _compressed_ message (it was compressed by {{COMPRESS}} on the first send). {{COMPRESS}} will try to compress the message again. To prevent this, we could check if the message has a {{CompressHeader}} and skip compression if so. However, small messages, or messages which grew in size rather than shrunk, will not have a header, so perhaps we need to add a {{CompressHeader}} in all instances and add a flag which tells {{COMPRESS}} whether to uncompress or not. was (Author: belaban): OK, I'll take a look whether byte[] arrays can be reused, too. Something to try out might be to replace the payload of the existing messages *in-place* and add a {{CompressHeader}} if successful. We have to consider the following cases: h5. {{COMPRESS}} is _above_ {{NAKACK2}} / {{UNICAST3}} The message is added to {{NAKACK2}} with a compressed payload and a {{CompressHeader}} attached. NAKACK2 will retransmit the _compressed_ message h5. {{COMPRESS}} is _below_ {{NAKACK2}} / {{UNICAST3}} NAKACK2 will retransmit the _compressed_ message (it was compressed by COMPRESS on the first send). COMPRESS will try to compress the message again. To prevent this, we could check if the message has a CompressHeader and skip compression if so. However, small messages, or messages which grew in size rather than shrunk, will not have a header, so perhaps we need to add a CompressHeader in all instances and add a flag which tells COMPRESS whether to uncompress or not. -- This message was sent by Atlassian JIRA (v6.4.11#64026)