SPECsfs97_R1.v3 Result

Performance

Throughput (ops/sec) Response (msec) 3132 0.6 6226 0.9 9391 1.2 12553 1.4 15740 1.6 18804 1.7 22095 2.1 25243 2.4 28426 3.0 30452 4.5

Configuration

LG Type	LG1
LG Model	NEC Express5800/120Rb-1
Number and Type Processors	1 GHz Pentium III
Memory Size	1 GB
Operating System	Red Hat Linux 7.2, Linux Kernel 2.4.24
Compiler	used pre-compiled binaries
Compiler Options	N/A
Network Type	Intel Pro/1000 Network Connection

LG Type	LG2
LG Model	NEC Express5800/120Rd-1
Number and Type Processors	2.4 GHz Xeon
Memory Size	1 GB
Operating System	Red Hat Linux 7.2, Linux Kernel 2.4.24
Compiler	used pre-compiled binaries
Compiler Options	N/A
Network Type	Intel Pro/1000 Network Connection

Testbed Configuration
LG #	LG Type	Network	Target File Systems	Notes
1,2,3,4,5,6,7,8	LG1	N1	F1, F2	N/A
9,10	LG2	N1	F1, F2	N/A

Other Notes

• The system was a two-node active-active cluster housed in a single cabinet.
• Each node had an array controller and activated two FC loops in total.
• Each node was connected to the shared nine disk enclosures.
• Nine disk encolosures separated two groups. Each group was comprised of four or five disk encolosures.
• Each enclosure group connected each FC loop.
• The first disk enclosure, which joins first group, contained four Operating System disks for both nodes.
• Each of the other eight disk enclosures contained fourteen user disks (formed a RAID 1+0 group) and a spare disk.
• A filesystem was comprised of the four disk enclosures.
• 2 GB of the 4 GB main memory is battery backed, called NVRAM, and contains all data that is pending change on the disks. Specifically, the other 2 GB of main memory can be lost in a reboot or power loss failure without loss of file system data.
• Using the UPS, the contents of the NVRAM is saved into a reserved area of the OS disks within a limited time.
• The TOE handles TCP. Jumbo Ethernet frames are not used.
• The following NFS and filesystem options were used; maximum journal size, no quota, no atime, no subtreecheck, no root squash, no privileged port check.
• In the event of system crash, all data to be written back to disk is held in the portion of memory that is battery backed memory (NVRAM). This NVRAM protects pending data changes from loss during a short power interruption and/or system crash and reboot.
• In the event of short power interruption, the NVRAM memory controller goes into a self refresh mode, so that it preserves its data. The rebooting system checks if there is pending changes in the NVRAM area and completes this disk changes during a recovery phase.
• The battery lasts only a few hours. The SPEC requires 72hour data protection. Our system is equipped with UPS. The UPS has enough energy to keep the controller and the first disk enclosure operational for 15 minutes. The UPS detects power failure and signal the system with the SNMP protocol.
• Once the system detects power failure, it goes into the special mode which is similar to the panic dump. Like panic dump, it stops any further I/O service and dump the NVRAM area into the special partition in the OS disk.
• The OS disk is mirrored. The (maximum of 2GB) NVRAM is saved sequentially to the OS disk within 10 minutes during which the UPS is guaranteed to provide power.
• At the next reboot with power coming back, the system checks if the data is saved in the special area and runs the special restoring program automatically. The program loads the special OS disk partition and restores the NVRAM area, before starting the usual OS. The usual OS sees the same NVRAM area as the system crash.

Generated on Thu Sep 2 14:26:40 2004 by SPEC SFS97 HTML Formatter
Copyright © 1997-2004 Standard Performance Evaluation Corporation