SPECsfs97_R1.v3 Result

Performance

Throughput (ops/sec) Response (msec) 20128 0.6 40354 0.8 60815 0.8 81059 0.8 101596 0.9 121801 1.1 142507 1.2 162751 1.5 183032 1.8 203182 2.6

Configuration

LG Type	LG1
LG Model	SuperServer 5013C-T
Number and Type Processors	1x2.8 GHz Pentium 4, 1MB cache
Memory Size	1024 MB
Operating System	Linux 2.4
Compiler	gcc 3.2
Compiler Options	-g -DNO_T_TYPES -DUSE_INTTYPES -DLinux
Network Type	Intel e1000 Gigabit Ethernet, MTU = 9000

Testbed Configuration
LG #	LG Type	Network	Target File Systems	Notes
1/2/../24	LG1	N1	F1	N/A

Other Notes

• The system under test is an ExaStore EX600FC which is a six node cluster with six 56-disk LSI 2882 FC Storage Systems.
• The cluster is comprised of 3 redundant node pairs. Each pair includes two nodes and two 2882 FC Storage Systems.
• Each 2882 FC Storage System contains one 2882 array module which contains two 2882 controllers.
• Each 2882 controller has a single connection to a different FC adapter on each of the pair nodes.
• Each 2882 controller has 1 GB of battery-backed RAM for a total of 1 GB of storage controller NVRAM.
• NVRAM cache mirroring was disabled in the 2882 controller.
• Read ahead was disabled in the 2882 controller.
• Each 2882 array module holds 14 disks. Each 2882 FC Storage System has three 2 Gb FC Drive Modules.
• Each 2 Gb FC Drive Module holds 14 disks.
• The 2882 array module and the 2 Gb FC Drive Modules are chained together with redundant FC-ALs.
• The single file system, F1, was comprised of 24 RAID-5 sets.
• Each RAID set had a strip size of 256 KB.
•  
• Uniform Access Rules (UAR):
• Each Node in the cluster has two N1 network ports connected to the server network (N1) for communication
• with all Load Generators.
• Each node in the cluster has two N2 network ports connected to the cluster network (N2) via 2 trunked
• Extreme Summit1i switches for communication with the other nodes.
• Each node in the cluster access 1/6 of the 2882 FC Storage Systems RAID sets.
• Each load generator process accessed the cluster through a single node and had full access to all the data.
• Each of the 2 nodes served as the access point for 1/6 of the load generator processes.
• Each node reads and writes 1/6 of its dataset to its attached RAID sets;
• the other 5/6 are read and written via the cluster network (N2).
•  
• Stable Storage:
• All memory in the ExaStore was used by the system for general-purpose memory and dynamically sized data cache.
• There was an integral UPS for each ExaStore node, which guarantees sufficient power in the event of power failure
• to dump the write cache to its local SCSI RAID. In case of unsafe situations (e.g. low battery condition),
• the system uses a journal mechanism to synchronously log all write-type operations to the LSI 2882 FC Storage Systems
• which are connected to all nodes. When power is restored, the ExaStore system is able to recover all committed data
• and metadata operations from journal files or dump files accordingly.
•  
• Server Tunings:
• The nodes read ahead option was disabled.
• The nodes write cache mirroring was disabled. This option enables mirroring of newly written data to a peer node to
• support a higher standard of high availability in addition to the UPS backed memory.
• All periodical system diagnostics and monitoring jobs were disabled for reproducibility:
• The cluster failover manager was disabled.
• Logging of periodical statistics was disabled.
•  
• Event log compression and mail delivery were disabled.
•  
• Exanet and ExaMesh are trademarks of Exanet, Inc.

Generated on Wed Oct 13 14:29:53 2004 by SPEC SFS97 HTML Formatter
Copyright © 1997-2004 Standard Performance Evaluation Corporation

First published at SPEC.org on 13-Oct-2004