click to visit home page
leading the way to the new storage frontier .....
SSD ad - click for more info
SSD news
what's the state of DWPD?
this way to the petabyte SSD
11 Key Symmetries in SSD design
where are we now with SSD software?
the Survivor's Guide to Enterprise SSDs
7 SSDs satisfy all future enterprise needs
4 things which changed in SSD year 2015
big vs small in flash array controller architecture
Why can't SSD's true believers agree about architecture?
where are we heading with memory intensive systems and software?

Editor:- although Auspex finally joined our list of dead storage companies (in June 2003) this classic article below remains a lasting testament to their forward looking technical vision. They were the first true NAS company, before the term "NAS" had even been invented. The ideas they developed about network storage architecture will continue in use long after the company itself has been forgotten.

... Auspex Systems - click to see profile

A Storage Architecture Guide

classic white paper by:- Auspex Systems - May 2000
Pure Storage was the #1 most researched SSD company.
Top 25 SSD Companies - 2016 Q2

High end solo enterprise PCIe SSDs are under threat of being squeezed out in the gap between DIMM wars memories and SATA SSDs.
the changing shape of the enterprise PCIe SSD market

90% of the enterprise SSD companies which you know have no good reasons to survive.
this way to market consolidation

SSD ad - click for more info

"A single petabyte of enterprise SSD could replace 10 to 50 petabytes of raw HDD storage in the enterprise - and still run all the apps faster and at lower cost - due to storage software and architecture efficiencies."
meet Ken and the SSD event horizon

SSD ad - click for more info

Chapter 1 - The importance of an effective storage strategy

With network content expected to continue expanding at explosive rates over the next 5 years, analysts predict that enterprise storage will account for 75% of all computer hardware expenditures. Such expansion makes it increasingly important for IT professionals to develop comprehensive strategies designed to optimize network infrastructure with storage solutions that will enable scalability, reliability, performance, availability, affordability and manageability.

Compounding the challenge of explosive content demands, there are two major technology shifts IT professionals must consider when developing an enterprise storage strategy:
  • first, the impact of networking technology on storage architecture and content management
  • second, the impact of parallel processing on the design of storage products.
These two technology shifts have produced three mutually coexistent methods for connecting storage to computing platforms: Direct Attached Storage (DAS), Network Attached Storage (NAS), and Storage Area Networks (SAN). Auspex Storage Architecture Guide is designed to help CEOs, CIOs and network adminis-trators understand these architectures and know the best use for each, while it provides suggestions for designing an effective storage strategy for workgroup, enterprise and ebusiness.
Storage standards are weak standards that are driven by component considerations. Network standards are strong standards that are driven by system considerations

A weak standard for Direct Attached Storage (DAS)

Today, greater than 95% of all computer storage devices such as disk drives, disk arrays and RAID systems are directly attached to a client computer through various adapters with standardized software protocols such as SCSI, Fibre Channel and others. This type of storage is alternatively called captive storage, server attached storage or direct attached storage (DAS) as illustrated in Figure 1.

The committees that established these standards, however, allowed such wide flexibility in interoperability that there a many variations of SCSI and Fibre Channel (FC) for the many available UNIX and Windows NT systems. For example, there are seven variations of SCSI, and most UNIX vendor implements FC differently. This is because storage was local to a specific server when these standards were defined and server vendors implemented variations that were not compatible. Storage standards therefore are weak standards and driven by component considerations. In other words, the problem with storage standards is that there seems to be so many of them.

As a result of weak storage standards, third-party DAS vendors such as EMC and Compaq Corporation, need to re-qualify their products with each revision of a server's operating system software. This can often lead to long lists of supported operating systems for SCSI or FC interconnects to different hosts. Each interconnect often requires special host software, special firmware and complicated installation procedures.
Figure 1. Direct Attached Storage (DAS) topology
A strong standard for Network Attached Storage (NAS)

In contrast, network standards are strong standards that are driven by system considerations. There are two true network standards for accessing remote data that have been broadly implemented by virtually all UNIX and Windows NT system vendors.
  • Developed and put into the public domain by Sun Microsystems, Network File System (NFS) is the de-facto standard for UNIX.
  • Developed by IBM and Microsoft, Common Internet File System (CIFS) is the standard for all flavors of the Windows operating system.

As a result of these broadly accepted standards for network data access, storage devices that serve data directly over a network (called Network Attached Storage or NAS devices) are far easier to connect and manage than DAS devices. Also, NAS devices support true file sharing between NFS and CIFS computers, which together account for the vast majority of all computers sold (see Figure 2).
Figure 2. Network Attached Storage (NAS) topology
Networks are now faster than storage channels

During the past five years the transfer rate for leading edge Direct Attached Storage (DAS) interconnects has increased fivefold from 20MB per second for F/W SCSI-2 to 100MB per second for Fibre Channel. Over this same period, however, the transfer rate for leading edge networking interconnects has increased tenfold from 12.5MB per second for 100baseT Ethernet to 128MB per second for Gigabit Ethernet. In other words, network data rates have not only caught up, but have surpassed direct attached storage (DAS), and are no longer two times slower — as they were five years ago. This has shifted the bottleneck from the network to the server and its direct attached storage.
Analysts predict a major shift from DAS to SAN and NAS

As DAS vendors were involved in the never-ending task of supporting all flavors of UNIX, NT, SCSI and FC for their storage products, both Dataquest and IDC recently began projecting explosive growth for NAS and SAN products as a percentage of the total storage market. These projections are based on four key factors:
  1. Strong standards for NAS result in simpler installation and lower management cost.
  2. Increased network speed can equalize the performance gap that used to exist between NAS and DAS for many applications.
  3. True data sharing between heterogeneous clients is possible with NAS and not with DAS.
  4. Trends to re-centralize storage to reduce management costs.

Dataquest has predicted that the commanding DAS 95% storage market share of today will be eclipsed by NAS over the next five years, and IDC projects that specifically designed NAS products will grow fivefold, from $1 billion in 1998 to $5 billion by 2002.
Scale-up - refers to architecture that uses a fixed controller resource for all processing. Scaling capacity happens by adding storage shelves up to the maximum number permitted for that controller.

Scale-out - refers to architecture that doesn't rely on a single controller and scales by adding processing power coupled with additional storage.
the Scale-Out vs. Scale-Up in enterprise flash arrays - (architecture ABCs) - a blog by Shachar Fienblit, CTO - Kaminario (May 2014)
Current confusion over NAS and SAN

Server vendors have implemented a variety of specialized hardware and software schemes to encourage the sale of storage with their processors. General-purpose DAS vendors have followed the same strategy. Not wanting to support NAS, where it would be easier for competitors to make inroads due to the clear NFS/CIFS standards, general purpose server vendors and general purpose storage vendors have developed their own proprietary visions of network storage. These visions are alternatively called Storage Networks (SNs) or Storage Area Networks (SANs).

The vendors developed these proprietary visions to bring the benefits of NAS to their users without losing control of the storage and networking sale to NAS vendors. The SAN initiative is a loose configuration of vendors attempting to promulgate the weak standards of the past while talking about bringing the benefits of networking to storage architecture. Benefits that are available with NAS, but that are still considered futures for SAN are as follows:
  1. LAN and server-free backup.
  2. 2. Storage resource pooling/sharing.
  3. Easy storage resource management.
  4. Data sharing.
  5. Interoperability of heterogeneous servers and storage.
applications aware DRAM, compression techniques in caches, power saving referesh, and a few other ideas
Are you ready to rethink enterprise RAM?
Instead of putting the storage directly on the network, the emerging SAN concept puts a network in between the storage subsystems and the server as (Figure 3). This means that SAN actually adds network latency to the DAS storage model. SAN standards are in forma-tive stages and may not be established for years. But, leading storage vendors have announced proprietary SANs that are still largely visions.

EMC has announced a proprietary Enterprise Storage Network (ESN), and Compaq has announced a proprietary Enterprise Network Storage Architecture (ENSA). As with UNIX and SCSI, SAN is likely to become a variety of similar architectures that are not based on strong standards. This may create major road-blocks to successful integration and data sharing between heterogeneous platforms. Both NAS and SAN are valid technologies and serve important roles with different objectives. However, because of the complexity arising from the many varieties of SCSI, UNIX, and proprietary SANs, a small percentage of storage today is actually connected to SANs. In a recent survey of UNIX and NT sites with over 5,000 employees by ITcentrix Inc., only 7% of enterprises have actually implemented SAN in production compared to about 48% that have implemented NAS.

The importance of parallel processing in storage subsystem design

Equal in importance to the impact of networking technology on storage architecture and management is the shift to parallel-processing architectures in storage subsystem design. Experts have noted the semiconductor industry finds it increasingly difficult to achieve faster processing speeds. To counteract this situation, storage subsystem providers are taking advantage of parallel processing designs in two ways:
  1. Designing computer nodes with multiple CPUs.
  2. Linking multiple nodes together to act as one system.

Among available NAS designs, only the Auspex 4Front NS2000 (Auspex NetServer 2000) series of content servers uses both approaches, making it the most advanced NAS product design available. In this design, each of three I/O nodes has two Intel processors. Each processor runs specialized real-time software called the DataXpress kernel. One I/O node processor, called the Network Processor (NP), manages highly reliable customized software that controls all network protocol and caching functions. The other I/O node processor, called the File and Storage Processor or FSP, handles file system processing and storage processing. This design, called Functional Multiprocessing (FMP), allows network processing for a second I/O to occur in parallel with data retrieval for the first I/O. The FMP design provides advantages over both nonparallel single processor and Symmetric Multipro-cessing (SMP) designs that handle I/O activity in serial. These advantages are discussed further in Chapter 5.
Figure 3. Storage Area Network (SAN) topology is still largely vision
This modern parallel architecture allows the Auspex NetServer 2000 to provide users with simplified scaleup of storage capacity, processors, network connections and perfor-mance. Also, this design provides major advantages in reducing backup windows due to parallel backup of data on each I/O node, as well as the ability to perform parallel backups across three nodes. Computing functions for efficient parallel processing of I/O are distributed through multiple busses in the I/O and host nodes of the Auspex NetServer 2000.

In comparison, Network Appliance relies on a single CPU architecture first designed in 1995 for their NAS product line. The Network Appliance design requires all storage requests to arbitrate for a single data bus and a single processor performs all computing functions. EMC Celerra provides a degree of parallelism by clustering network adapters (known as Data Movers) in a separate cabinet in front of the Symmetrix storage subsystem and by removing data management functions to separate processors, known as Control Stations. Like the Network Appliance F700 series, all storage requests from the Celerra network adapters must arbitrate for a single redundant backplane bus in the Symmetrix storage subsystem that processes all I/O serially.

Auspex is committed to providing the best possible NAS information

Being the primary innovator of NAS, Auspex is widely considered by customers and analysts to have the highest level of expertise in both NAS and networking technology. Auspex believes that there is a valid role for NAS, DAS and SAN in enterprise storage architecture strategy. Since the topic of NAS is new to many customers, Auspex is committed to providing the best available information about optimizing these different types of storage depending for real application work-loads, cost, performance, availability and data sharing. This report contains URLs and references that identify some of the available information. Current information is also regularly published at
The Network Appliance design requires all storage requests to arbitrate for a single data bus and a single processor performs all computing functions

All storage requests from the Celerra network adapters must arbitrate for a single redundant backplane bus in the Symmetrix storage sub-system
Drilling to the details

Chapter 1 has identified trends in computing that affect network storage strategy decisions, introduced the main storage architecture options, and identified issues involved in selecting an appropriate enterprise storage solution. Chapter 2 defines the different types of storage architecture, comparing benefits, technologies and applications. Chapter 3 provides guidelines on the best use for each storage architecture. Chapter 4 compares the subsystem architecture used by each major enterprise storage vendor. Chapter 5 discusses the benefits of implementing a parallel hardware and software design in enterprise storage strategy. Chapter 6 brings the pieces together in a conclusion designed to help CIOs and network administrators identify the best total solution for their specific enterprise storage needs. The best total solution goes beyond selecting just the right storage architecture and product, to identifying the vendor who offers the most advanced technology, has a proven history of providing innovative enterprise storage systems and solutions, and who provides value-added service, consulting, and support.

Chapter 2 - DAS, NAS and SAN storage architectures defined

Direct Attached Storage

DAS evolved from the server industry where server vendors have traditionally sold storage as an add-on. DAS is an appropriate choice for the following: Very low end PC applications Very high-end high-performance mainframe applications Certain compute intensive and high performance OLTP database applications.
Direct Attached Storage network topology
Network Attached Storage

NAS evolved from the networking industry where there are strong standards for connectivity, data security and load balancing. NAS is a more mature technology and deployed in 1 of 2 enterprises today.

NAS is the best choice for the following types of applications: UNIX and Windows NT data sharing applications Consolidated file serving applications Technical and scientific applications Internet and Intranet applications e-Business and certain types of Decision Support (DSS) applications
Network Attached Storage (NAS) topology
Storage Attached Network

SAN evolved from the storage industry where standards have been weak, therefore SAN implementations are not standardized. SAN is deployed in 1 of 20 enterprises. Because of its current lack of standards SAN is available only in proprietary configurations and the long-term inter-operability of these schemes is still not apparent. Although the SAN vision involves many benefits now available on NAS, there are certain synergies with existing enterprise operational and management tools that have led early adopters to experiment with test deployments.

Although the long-term SAN vision is for inter-operability among heterogeneous servers and storage products, it is advisable to implement early SAN applications in a homogeneous environment with one of the available proprietary solutions from EMC, HDS or Compaq. SAN is also appropriate for applications that do not require true data sharing, a feature which will not likely be available until SAN standards evolve to the level of those available with NAS today. SAN is also appropriate for applications where the well known Fibre Channel (FC) and Fibre Channel Protocol for SCSI (FCP) security risks can be managed and where performance bottlenecks arising form Fibre Channel node and link congestion can be avoided.
Storage Attached Network network topology
Making the right product choice for serving content

Although analysts have predicted a major shift from DAS to NAS over the next five years, many enterprises still use general-purpose servers to serve files in UNIX and Windows NT environments. This choice, however, has drawbacks in the areas of availability, performance, scalability, and manageability compared to products from Network Appliance, EMC and Auspex which are more suited for dedicated network file serving. Following is a brief discussion of the enterprise storage solutions available from these three vendors:
Network Appliance F700: Single process, single system bus

The single processor, single system bus, single parity disk, RAID 4 Network Appliance F700 Series was designed for low-end, low-cost, non-mission critical applications for which it is an appropriate choice.

Network Appliance is likely to announce an Intel based SMP design to allow greater scalability and performance than its current product line. Application upgrade capability will be an important issue to consider when this new product is announced.
The "single processor/single system bus" design of the Network Appliance F700 Series.
EMC Celerra: clustered network

EMC's Celerra provides a clustered network file server front end to the Symmetrix storage subsystem which was originally designed for mainframe data. Celerra and Symmetrix is appropriate for co-located mainframe and open system data and for users requiring remote mirroring for disaster tolerance.

The Celerra hybrid design has certain restrictions compared to Network Appliance or Auspex when it comes to NFS and CIFS data sharing and should be viewed more as partitioned storage than as an optimized file serving, data sharing product.
EMC Celera clustered network system diagram
Auspex 4Front NS2000: Functional Multiprocessing (FMP)

Among available products that are specifically designed for file serving the Auspex 4Front NS2000 Series is the only choice specifically designed for file serving that is based on a parallel processing design using integrated hardware and software.

The NS2000 architecture is known as the Functional Multiprocessing (FMP) architecture and has been patented by Auspex. FMP is based on a building block concept that is highly scalable and easy to expand. Auspex chose an FMP over an SMP design because an FMP design distributes different functions to the multiple CPUs in each I/O node. In addition, the Auspex design distributes different functions to the host node compared to the I/O node and allows multiple I/O nodes to function in parallel.
Auspex 4Front NS2000: Functional Multiprocessing (FMP)

The above extract is a small part of a comprehensive white paper from Auspex Systems. Our thanks again to Rob Killen in Auspex UK for giving us permission to reproduce it. We hope you've found it thought provoking.

Click here to view the full 50 page article - a Storage Architecture Guide (pdf)

The above mentioned pdf was an updated 2nd edition of the original guide - published a year later - in October 2001.

Compared to EMC...
ours is better

can you take these AFA companies seriously?


storage search banner 1.0" SSDs 1.8" SSDs 2.5" SSDs 3.5" SSDs PCIe SSDs rackmount SSDs

STORAGEsearch is published by ACSL