NAS appliances have traditionally attained clustered redundancy through the use of NUMA implementations where synchronous mirrors are maintained over a private network or local interconnect.

Figure 1

This approach has been successful in its simplicity and real-world reliability. However, drawbacks have also been significant. These include:
 

 

  • Performance
    Performance penalties from maintaining a synchronous mirror over a cable interconnect, wherein client requests are not completed until the mirror partner has logged that request and sent a completion status to the primary server.

    A side effect of this architectural limitation is that it can lead to vendors proposing, or customers creating, costly private networks based on custom or exotic interconnects (VI fibre channel, Infiniband, Myrinet, etc.) in an effort to reduce this latency from the synchronous mirror. In addition to being expensive, these networks are challenging to support and tend to have a limited mindshare/knowledge base.

  • Scalability
    A lack of scalability, meaning that a performance or capacity requirement for a single additional NAS appliance would result in the need to purchase two appliances to maintain clustered redundancy.

  • Availability
    A lack of true high availability. Despite the excellent real-world reliability experienced from NUMA style clustered NAS appliances, today’s goals for true high availability require more than x1 redundancy. This is not easily possible when maintaining a synchronous mirror, as it would resort to adding more partners and significantly more overhead than a two-way mirror.

  

Micro Memory believes it has solved these issues for clustered storage appliances with its MM-5460CN that includes completely embedded, battery backed power for its on-board PowerPC processor, Gigabit Ethernet, PCI Mezzanine (PMC) Site, and SDRAM memory array.

 

<1  2  3  4>