Fix indentation levels.

Sponsored by:	The FreeBSD Foundation
This commit is contained in:
Glen Barber 2014-07-11 15:01:08 +00:00
parent 52914213d1
commit 8cd45d8b40
Notes: svn2git 2020-12-08 03:00:23 +00:00
svn path=/head/; revision=45252

View file

@ -123,50 +123,49 @@
</project> </project>
<project cat='proj'> <project cat='proj'>
<title>RPC/NFS and CTL/iSCSI performance optimizations.</title> <title>RPC/NFS and CTL/iSCSI performance optimizations.</title>
<contact> <contact>
<person> <person>
<name> <name>
<given>Alexander</given> <given>Alexander</given>
<common>Motin</common> <common>Motin</common>
</name> </name>
<email>mav@FreeBSD.org</email> <email>mav@FreeBSD.org</email>
</person> </person>
</contact> </contact>
<body> <body>
<p>The &os; RPC stack, used as base for its NFS server, took <p>The &os; RPC stack, used as base for its NFS server, took
multiple optimizations to improve its performance and SMP multiple optimizations to improve its performance and SMP
scalability. Algorithmic optimizations allowed to reduce scalability. Algorithmic optimizations allowed to reduce
processing overhead, while improved locking allowed it to processing overhead, while improved locking allowed it to
scale up to at least 40 processor cores without significant scale up to at least 40 processor cores without significant
lock congestion. Combined with some other kernel lock congestion. Combined with some other kernel
optimizations that allowed to increase peak NFS request optimizations that allowed to increase peak NFS request rate
rate by many times, reaching up to 600K requests per second by many times, reaching up to 600K requests per second on
on modern hardware.</p> modern hardware.</p>
<p>The CAM Target Layer (CTL), used as base for new kernel <p>The CAM Target Layer (CTL), used as base for new kernel iSCSI
iSCSI server, also took series of locking optimization, server, also took series of locking optimization, that allowed
that allowed to increase its peak request rate from ~200K to increase its peak request rate from ~200K to ~600K IOPS
to ~600K IOPS with potential of reaching reate of 1M with potential of reaching reate of 1M request per second.
request per second. That rate is sufficient to completely That rate is sufficient to completely saturage 2x10Gbit
saturage 2x10Gbit Ethernet links with 4KB requests. For Ethernet links with 4KB requests. For comparison, the port of
comparison, the port of net/istgt (user-level iSCSI net/istgt (user-level iSCSI server) on the same hardware with
server) on the same hardware with equal configuration equal configuration shown only 100K IOPS.</p>
shown only 100K IOPS.</p>
<p>There is also ongoing work on improving CTL functionality. <p>There is also ongoing work on improving CTL functionality.
It was already made to support 3 of 4 VMWare VAAI storage It was already made to support 3 of 4 VMWare VAAI storage
acceleration primitives (net/istgt supports 2), while the acceleration primitives (net/istgt supports 2), while the goal
goal is to reach full VAAI support during next months.</p> is to reach full VAAI support during next months.</p>
<p>With all above, and earlier improvements in CAM, GEOM, ZFS <p>With all above, and earlier improvements in CAM, GEOM, ZFS
and number of other kernel areas coming soon FreeBSD 10.1 and number of other kernel areas coming soon FreeBSD 10.1 may
may become the fastest storage release ever. ;)</p> become the fastest storage release ever. ;)</p>
<p>These projects are sponsored by iXsystems, Inc.</p> <p>These projects are sponsored by iXsystems, Inc.</p>
</body> </body>
</project> </project>
<project cat="arch"> <project cat="arch">