SPECsfs2008_nfs.v3 Result

Throughput (ops/sec) Response (msec) 7894 0.8 15814 1.2 23752 1.5 31663 1.9 39610 1.3 47541 1.7 55538 3.0 63752 4.0 71771 5.0 77137 7.3

Tested By	Panasas, Inc.
Product Name	Panasas ActiveStor Series 9
Hardware Available	October, 2009
Software Available	October, 2009
Date Tested	December, 2009
SFS License Number	3866
Licensee Locations	Fremont, California, USA

The Panasas Series 9 NAS features a parallel clustered file system that turns files into smart data objects, and dynamically distributes and load balances data transfer operations across a networked blade architecture. The Panasas ActiveScale distributed file system creates a storage cluster with a single file system and single global namespace. Each self-contained shelf chassis houses up to eleven DirectorBlade and StorageBlade modules, two redundant network switch modules, two redundant power supplies, and a battery backup module. All components are hot-swappable.

As configured, each shelf includes 8 TB HDD storage, 256 GB SSD storage, 44 GB RAM, and dual 10 GbE network connectivity with the network failover option. Systems can be configured with any number of shelves to scale capacity and performance, in a single global namespace, with a single point of management.

Item No	Qty	Type	Vendor	Model/Name	Description
1	10	Panasas Storage Shelf	Panasas, Inc.	900027-004	PAS 9, 3+8, 8TB + 256GB SSD, 32GB Cache, 10GE SFP+

OS Name and Version	ActiveScale 3.5.0.c
Other Software	None
Filesystem Software	ActiveScale 3.5.0.c

Name	Value	Description
None	None	None

Server Tuning Notes

None

Description	Number of Disks	Usable Size
Each DirectorBlade has one 80 GB SATA 7200 RPM HDD that holds the ActiveScale OS, but not filesystem metadata or user data.	30	2.3 TB
Each StorageBlade has one 1000 GB SATA 7200 RPM HDD that contains the ActiveScale OS, filesystem metadata, and user data.	80	78.1 TB
Each StorageBlade has one 32 GB SSD that contains filesystem metadata and user data.	80	2.5 TB
Total	190	83.0 TB

Number of Filesystems	1
Total Exported Capacity	74840 GB
Filesystem Type	PanFS
Filesystem Creation Options	Default
Filesystem Config	The 10 shelves (80 StorageBlades) in the SUT were configured in a single bladeset, which is a fault-tolerant shared pool of disks. All 30 volumes in the file system share space in this bladeset. Each volume stripes data uniformly across all disks in the pool using Object RAID 1/5, which dynamically selects RAID-1 (mirroring) or RAID-5 (XOR parity) on a per-file basis. The bladeset was configured with vertical parity enabled. Vertical parity corrects media errors at the StorageBlade level, before they are exposed to Object RAID, and provides protection equivalent to RAID-6 against a sector error during a RAID rebuild.
Fileset Size	9256.5 GB

Item No	Network Type	Number of Ports Used	Notes
1	Jumbo 10 Gigabit Ethernet	10	None

Network Configuration Notes

Each Panasas shelf used a single 10 Gigabit Ethernet interface configured with MTU=9000 (jumbo frames). These links were connected to the Cisco Nexus 5020.

Four "Duo Node" ASUS servers (comprising eight nodes) and seven Penguin nodes, for a total of 15 client nodes, were used as load generators and connected via 10 GbE NICs to the Cisco Nexus 5020 switch. The 10 PAS9 shelves were also connected to the same Nexus 5020. There were no additional settings on the clients or the shelves, and the 25 ports were configured in a single VLAN on the Nexus 5020.

Item No	Qty	Type	Description	Processing Function
1	30	CPU	Pentium-M 1.8 GHz	NFS Gateway, Metadata Server, Cluster Management, iSCSI, TCP/IP
2	80	CPU	Celeron-M 1.5 GHz	Object Storage File System, iSCSI, TCP/IP

Processing Element Notes

The SUT includes 30 DirectorBlades, each with a single Pentium-M 1.8 GHz CPU, and 80 StorageBlades, each with a single Celeron-M 1.5 GHz CPU. The DirectorBlades manage filesystem metadata and provide NFS gateway services. The StorageBlades store user data and metadata, and provide access to it through the OSD (Object Storage Device) protocol over iSCSI and TCP/IP. One DirectorBlade in the SUT was also running the Panasas realm manager and management user interface.

Description	Size in GB	Number of Instances	Total GB	Nonvolatile
Blade main memory	4	110	440	NV
Grand Total Memory Gigabytes			440

Memory Notes

Each DirectorBlade and each StorageBlade has 4 GB ECC RAM.

Each shelf has two (redundant) power supplies as well as a battery, which powers the entire shelf for about five minutes in the event of AC power loss. In the event of a power failure, each blade saves cached writes from main memory to its local HDD before shutting down. The data saved in this way is maintained indefinitely while the system is powered down, and is automatically recovered when power is restored.

The SUT is comprised of 10 Panasas PAS9 shelves, with each shelf connected via 10 GbE SFP+ to a Cisco Nexus 5020 10 GbE switch. Each of the 15 load generators were connected to the Cisco Nexus 5020 with 10 GbE SFP+.

Item No	Qty	Vendor	Model/Name	Description
1	4	ASUS	RS700D-E6/PS8	ASUS "Duo Node" Linux server
2	7	Penguin Computing	Altus 1650SA	Penguin Linux server
3	1	Cisco	Nexus 5020	Cisco 40-port 10 GbE Switch

LG Type Name	ASUS
BOM Item #	1
Processor Name	Intel Xeon E5530
Processor Speed	2.4 GHz
Number of Processors (chips)	2
Number of Cores/Chip	4
Memory Size	1 GB
Operating System	CentOS 5.3 (2.6.18-128.el5)
Network Type	1 Intel 82598EB 10 Gigabit NIC

LG Type Name	Penguin
BOM Item #	2
Processor Name	AMD Opteron 2356
Processor Speed	2.3 GHz
Number of Processors (chips)	2
Number of Cores/Chip	4
Memory Size	1 GB
Operating System	CentOS 5.3 (2.6.18-128.el5)
Network Type	1 Intel 82598EB 10 Gigabit NIC

Benchmark Parameters

Network Attached Storage Type	NFS V3
Number of Load Generators	15
Number of Processes per LG	30
Biod Max Read Setting	2
Biod Max Write Setting	2
Block Size	AUTO

Testbed Configuration

LG No	LG Type	Network	Target Filesystems	Notes
1..1	ASUS	N1	F1 (GW1:/V16, GW2:/V16, ..., GW15:/V16; GW16:/V01, GW17:/V01, ..., GW30:/V01)
2..2	ASUS	N1	F1 (GW1:/V17, GW2:/V17, ..., GW15:/V17; GW16:/V02, GW17:/V02, ..., GW30:/V02)
3..3	ASUS	N1	F1 (GW1:/V18, GW2:/V18, ..., GW15:/V18; GW16:/V03, GW17:/V03, ..., GW30:/V03)
4..4	ASUS	N1	F1 (GW1:/V19, GW2:/V19, ..., GW15:/V19; GW16:/V04, GW17:/V04, ..., GW30:/V04)
5..5	ASUS	N1	F1 (GW1:/V20, GW2:/V20, ..., GW15:/V20; GW16:/V05, GW17:/V05, ..., GW30:/V05)
6..6	ASUS	N1	F1 (GW1:/V21, GW2:/V21, ..., GW15:/V21; GW16:/V06, GW17:/V06, ..., GW30:/V06)
7..7	ASUS	N1	F1 (GW1:/V22, GW2:/V22, ..., GW15:/V22; GW16:/V07, GW17:/V07, ..., GW30:/V07)
8..8	ASUS	N1	F1 (GW1:/V23, GW2:/V23, ..., GW15:/V23; GW16:/V08, GW17:/V08, ..., GW30:/V08)
9..9	Penguin	N1	F1 (GW1:/V24, GW2:/V24, ..., GW15:/V24; GW16:/V09, GW17:/V09, ..., GW30:/V09)
10..10	Penguin	N1	F1 (GW1:/V25, GW2:/V25, ..., GW15:/V25; GW16:/V10, GW17:/V10, ..., GW30:/V10)
11..11	Penguin	N1	F1 (GW1:/V26, GW2:/V26, ..., GW15:/V26; GW16:/V11, GW17:/V11, ..., GW30:/V11)
12..12	Penguin	N1	F1 (GW1:/V27, GW2:/V27, ..., GW15:/V27; GW16:/V12, GW17:/V12, ..., GW30:/V12)
13..13	Penguin	N1	F1 (GW1:/V28, GW2:/V28, ..., GW15:/V28; GW16:/V13, GW17:/V13, ..., GW30:/V13)
14..14	Penguin	N1	F1 (GW1:/V29, GW2:/V29, ..., GW15:/V29; GW16:/V14, GW17:/V14, ..., GW30:/V14)
15..15	Penguin	N1	F1 (GW1:/V30, GW2:/V30, ..., GW15:/V30; GW16:/V15, GW17:/V15, ..., GW30:/V15)

Load Generator Configuration Notes

Each of the 15 clients accessed all (30) gateways to 2 volumes (30 total) in the file system. In this way, all 30 volumes were accessed through all 30 gateways from these 15 clients. In all cases, each gateway accessed was serving a volume other than its primary volume, preventing any gateway/volume affinity in the file system.

Each Panasas DirectorBlade provides both NFS gateway services for the entire filesystem, and metadata management services for a portion of the filesystem (one or more virtual volumes). The system as tested contained 30 DirectorBlades and was configured with 30 volumes, one managed by each DirectorBlade.

To comply with the Uniform Access Rule (UAR) for a single namespace, each of the 15 clients accessed two volumes, and each volume was accessed through 15 NFS gateways. In all cases, the 15 NFS gateways used for a given volume were selected such that they did not contain the DirectorBlade providing metadata services for that volume. In this way, each client accessed all 30 gateways, all 30 volumes were accessed equally by the 15 clients, and no client was afforded any advantage by colocating the NFS gateway and metadata server for the volume being accessed.

The 80 Panasas StorageBlades were configured as a single bladeset (storage pool), so data from all 30 volumes was distributed uniformly across all StorageBlades. This results in uniform access from the clients to the storage, and assures that each server in the cluster serves the same amount of data traffic from each disk uniformly.

• Configuration