SPECsfs2008_nfs.v3 Result

Throughput (ops/sec) Response (msec) 7991 1.4 16005 1.7 24024 2.1 32039 2.4 40062 2.9 48108 3.4 56122 4.0 64124 5.0 72328 6.3 80279 10.0

Tested By	BlueArc Corporation
Product Name	BlueArc Mercury 50, Cluster
Hardware Available	July 2009
Software Available	October 2009
Date Tested	October 2009
SFS License Number	63
Licensee Locations	San Jose, CA, USA

The BlueArc Mercury Server is a next generation network storage platform that consolidates multiple applications and simplifies storage management for businesses with mid-range storage requirements. The Mercury platform leverages a hybrid-core architecture that employs both field programmable gate arrays (FPGAs) and traditional multi-core processors to deliver enhanced performance across a variety of application environments, including general purpose file systems, database, messaging and online fixed content.

The Mercury server series features 2 models: the Mercury 50 and the Mercury 100, which can scale to 1PB or 2PB of usable data storage respectively, supporting a range of SAS, FC, Nearline SAS and SATA drive technologies. Both servers support simultaneous 1GbE and 10GbE LAN speed access, and 4Gbps FC storage connectivity.

Item No	Qty	Type	Vendor	Model/Name	Description
1	2	Server	BlueArc	SX345260	BlueArc Mercury Server, Model 50
2	6	Disk Controller	BlueArc	SX345263	RC12C Storage Controller
3	6	Disk Expansion	BlueArc	SX345264	RC12Exp Storage Expansion
4	12	Disk Drive	BlueArc	SX345265	RC12 Disk 300GB SAS 15K Drive (12 Pk)
5	2	FC Switch	BlueArc	SX345120	FC Switch, 4Gbps 16-port w/SFPs
6	1	Server	BlueArc	SX345268	System Management Unit, Model 200
7	2	Software	BlueArc	SX435045	Software License (Tier B) - Unix (NFS)
8	2	Software	BlueArc	SX435053	Software License (Tier B) - Cluster
9	2	Software	BlueArc	SX435054	Software License (Tier B) - Cluster Namespace

OS Name and Version	SU 6.5.1848
Other Software	None
Filesystem Software	SiliconFS 6.5.1848

Name	Value	Description
cifs_auth	off	Disable CIFS security authorization
fsm cache-bias	small-files	Set metadata cache bias to small files
fs-accessed-time	off	Accessed time management was turned off
shortname	off	Disable shortname generation for CIFS clients
fsm read-ahead	0	Disable file read-ahead
security-mode	UNIX	Security mode is native UNIX

Server Tuning Notes

N/A

Description	Number of Disks	Usable Size
300GB SAS 15K RPM Disk Drives (Seagate ST3450856SS)	144	29.4 TB
160GB SATA 5400 RPM Disk Drives (Hitachi HTE543216L9A300) used for storing the core operating system and management logs. No cache or data storage.	4	320.0 GB
Total	148	29.7 TB

Number of Filesystems	2
Total Exported Capacity	20TB
Filesystem Type	WFS-2
Filesystem Creation Options	4K filesystem block size dsb-count (dynamic system block) set at 768
Filesystem Config	Each FS spans 9 LUNs of 6+2 RAID 6
Fileset Size	9373.7 GB

The storage configuration consisted of 6 RC12 RAID arrays, each with a single RC12C controller and a single RC12E expansion shelf for a total of 12 shelves of storage; each shelf with 12 disk drives. Each array was configured with three 6+2 RAID 6 LUNs.

The controller shelf has two 4Gbps FC H-ports connected to the server via a redundant pair of Brocade 200E switches. Each Mercury server is connected to each Brocade 200E via a 4 Gbps FC connection, such that a completely redundant path exists from each server to all storage. The expansion shelves connect to the controller by dual 6Gb/s SAS connections.

The two filesystems were aggregated into a single namespace ( /r ) using BlueArc.s Cluster Namespace (CNS) feature. One filesystem was assigned as the primary filesystem to each Mercury server, although both servers can access all filesystems through the active-active cluster link, or directly via the FC SAN in the event of a node failure.

Mercury servers have 2 internal mirrored hard disk drives per server which are used to store the core operating software and system logs. These drives are not used for cache space or for storing data.

Item No	Network Type	Number of Ports Used	Notes
1	10 Gigabit Ethernet	2	Integrated 1GbE / 10GbE Ethernet controller

Network Configuration Notes

One 10GbE network interface per Mercury server was used to connect to a Fujitsu XG2000 switch, which provided connectivity to the clients. The interfaces were configured to use jumbo frames (MTU size of 9000 bytes).

Each LG has a Myricom 10GbE single port PCIe network interface. Each LG connects via a single 10GbE connection to ports on the Fujitsu XG2000.

Item No	Qty	Type	Description	Processing Function
1	4	FPGA	Altera Stratix III EP3SE260	Storage Interface, Filesystem
2	4	FPGA	Altera Stratix III EP3SL340	Network Interface, NFS, Filesystem
3	2	CPU	Intel E8400 3.0GHz, Dual Core	Management
4	12	CPU	N/A	HW RAID

Processing Element Notes

Each Mercury server has 2 of each type of FPGA (4 total) used for the benchmark processing functions. Each RAID Controller shelf has dual controllers with 1 CPU each (2 total). There are 6 arrays used in this configuration.

Description	Size in GB	Number of Instances	Total GB	Nonvolatile
Server Main Memory	12	2	24	V
Server Filesystem and Storage Cache	14	2	28	V
Server Battery-backed NVRAM	2	2	4	NV
RAID Cache	1	6	6	V
Grand Total Memory Gigabytes			62

Memory Notes

Each Mercury server has 12GB of main memory that is used for the operating system and in support of the FPGA functions. 14GB of memory is dedicated to filesystem metadata and sector caches. A separate, integrated battery-backed NVRAM module on the filesystem board is used to provide stable storage for writes that have not yet been written to disk.

The RAID Controllers were configured in write-thru mode (write cache disabled) with read-ahead disabled.

The Mercury server writes first to the battery based (72 hours) NVRAM internal to the Mercury. Data from NVRAM is then written to the disk drives arrays at the earliest opportunity, but always within a few seconds of arrival in NVRAM. In the active-active cluster configuration, the contents of the NVRAM are synchronously mirrored to ensure that in the event of a single node failover, any pending transactions can be completed by the remaining node.

The system under test consisted of a two 3U Mercury 50 servers, connected to storage via 2 16-port 4Gbps FC switches. The servers are configured in an active-active cluster, directly connected by a redundant pair of 10GbE connections to the cluster interconnect ports. The storage consisted of 6 arrays of 24 drives each, with a RAID controller shelf and disk expansion shelf in each array. All connectivity from server to storage was via a 4Gbps switched FC fabric. For the purposes of the test, each FC switch was zoned into 2 zones, for a total of 4 distinct zones used in the test. Each Mercury 50 server connected to each zone via 4 integrated 4Gbps FC ports (corresponding to 4 H-ports) built into the server. Each storage array was connected to 2 zones (corresponding to 2 H-ports) providing a fully redundant path from each server to every storage array. Expansion shelves were connected to the controllers via dual SAS interconnects.

The System Management Unit (SMU) is part of the total system solution, but is used for management purposes only. It is not active during the test.

Item No	Qty	Vendor	Model/Name	Description
1	12	Unbranded	N/A	Unbranded 1U Solaris client with Tyan n6650W dual Opteron motherboard, 4GB RAM
2	1	Fujitsu	XG2000	Fujitsu XG2000 20-port 10GbE Switch

LG Type Name	LG1
BOM Item #	1
Processor Name	AMD Opteron 2218
Processor Speed	2.6 Ghz
Number of Processors (chips)	2
Number of Cores/Chip	2
Memory Size	8 GB
Operating System	Solaris 10
Network Type	1 x Myricom (10G-PCIE-8A-R+E) 10Gbit Single Port PCIe

Benchmark Parameters

Network Attached Storage Type	NFS V3
Number of Load Generators	12
Number of Processes per LG	64
Biod Max Read Setting	2
Biod Max Write Setting	2
Block Size	AUTO

Testbed Configuration

LG No	LG Type	Network	Target Filesystems	Notes
1..12	LG1	1	/r/f1, /r/f2	N/A

Load Generator Configuration Notes

All clients were connected to one common namespace on the server cluster, connected to a single 10GbE network.

Each load generating client hosted 64 processes, all accessing a single namespace on the Mercury cluster through a common network connection.

There are 2 physical target filesystems ( /r/f1 and /r/f2 ) that are presented as a single cluster namespace through virtual root ( /r ), accessible to all clients. Each load generator mounted to each filesystem target ( /r/f1 and /r/f2 ) and cycled through the two filesystems in sequence.

The filesystems were uniformly striped across the available storage arrays such that all clients uniformly accessed all arrays.

None

BlueArc and SilconFS are registered trademarks, and BlueArc Mercury is a trademark of BlueArc Corporation in the United States and other countries. All other trademarks belong to their respective owners and should be treated as such.

• Mercury 50 Cluster Test Configuration Diagram