Fujitsu SPARC T5440 User Manual page 88

Fixed Firmware and General Software Issues (Continued)
TABLE A-4
CR ID Description
6311743 Systems configured with multiple Ethernet
interfaces receiving large quantities of TCP data
might exhibit performance degradation with
large numbers of connections actively receiving
data. However, there is no system failure.
This issue will manifest itself on systems with
two or more 10 Gbit Ethernet interfaces, or more
than 8 Gbit Ethernet interfaces, with workloads
receiving high volumes of data over many
connections via TCP. In this situation, TCP
receive performance will not increase as
expected, and above a certain point the overall
receive performance will decrease.
6638112 Under heavy workloads with many threads or
processes simultaneously accessing the same
page of memory, the system might panic. The
panic message will indicate a send mondo
timeout
6640564 Guests in a logical domains (LDoms)
environment can exhibit poor virtual disk I/O
performance if the I/O domain is using a disk
image or an SVM volume instead of a physical
disk. This performance loss is particularly
visible if multiple applications are performing
I/O to the same virtual disk.
6693633 A multithreaded application with a hundred or
more threads that transmits a large number of
small (less than the MTU size) sized packets on
the network might experience performance
below that expected.
78 SPARC Enterprise T5440 Server Product Notes • December 2010
Workaround
Note - Fixed in Solaris patch ID 141414-07 or
later.
Check for the availability of a patch for this defect.
Note - Fixed in Solaris patch ID 141414-01 or
later.
Add the following line to the /etc/system file:
set xc_tick_limit_scale = 5
Reboot the system. If the panic recurs after
applying this workaround, contact your service
representative.
Note - Fixed in Solaris patch ID 139562-02 or
later.
Configure the I/O domain to use a physical disk
for the Guest domains' virtual disk. If a disk image
is required in the I/O domain, performance will
improve if the image is on a ZFS volume instead
of on an SVM volume.
Note - Fixed in Solaris patch ID 141444-03 or
later.
If you are not getting the performance you expect
and you are using a large number of CPUs or a
large number of threads in a virtual domain then
try using a smaller number of cpus or threads to
improve performance.