Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 5 Mar 2010 07:20:45 +0900
From:      "Daisuke Aoyama" <aoyama@peach.ne.jp>
To:        "Miroslav Lachman" <000.fbsd@quip.cz>, <freebsd-scsi@freebsd.org>
Cc:        ivoras@freebsd.org
Subject:   Re: performance tuning of iSCSI and Dell MD3000i / gjournal problem
Message-ID:  <96BBFBFE5763434E8F65D8C503BABB2C@artemis>
References:  4B5F1553.4070709@quip.cz <4B79DCA2.1070005@quip.cz>

next in thread | previous in thread | raw e-mail | index | archive | help
Hi,

> And ZFS on iSCSI is the worst case with 5MB/s write performance! (I have 
> much better experiences with ZFS on RAIDZ pool of 4 SATA drives)

I'm not sure but FreeBSD initiator does not support immediate operation. 
(ImmediateData=Yes)
If the target supports it, the initiator can send data with the command
up to the size in small one of FirstBurstLength and 
MaxRecvDataSegmentLength.
Also the initiator can send more data up to FirstBurstLength if the target
supports unsolicited operation. (InitialR2T=No)
It will improve small writing performance and latency.
In any case, each PDU must not exceed the target's MaxRecvDataSegmentLength.
See RFC3720 page 213 and 217 for example of the sequence.

http://tools.ietf.org/html/rfc3720#page-213
http://tools.ietf.org/html/rfc3720#page-217

If possible, try to increase MaxBurstLength both of the initiator and the 
target.
It will decrease the number of R2T. Therefore, large writing is faster.

Regards,
Daisuke Aoyama




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?96BBFBFE5763434E8F65D8C503BABB2C>