Stack Exchange Network

Stack Exchange network consists of 175 Q&A communities including Stack Overflow, the largest, most trusted online community for developers to learn, share their knowledge, and build their careers.

Visit Stack Exchange
5 of 6 clarified the meaning of the research by WISC by adding a related resource; deleting ZFS descriptions that are not relevant to this question

What is a non-critical server? One that can fail?

ECC RAM is fundamental when memory reliability is fundamental.

Two things grow with the growth of memory sizes:

  • the reliance of software on memory, esp. server software (take e.g. caching)
  • the probability of memory error (p = num_bits * p_bit_failure)

This intel presentation on ECC reports these facts:

  • Average rate of memory error for a server with 4GB memory running 24x7 is 150 times a year
  • ~4000 correctable errors per memory module per year
  • Overclocking and system age greatly increase failure rates
  • Recurrent failures are common and happen quickly (97% occur within 10 days of first failure) => avalanche effect
  • For an ECC server with lifespan of 3 to 5 years, chance for system failure uncorrectable memory error is less than 0.001%

Another recent research by WISC shows ECC to be essential for these ZFS systems:

ZFS has no precautions for memory corruptions: bad data blocks are returned to the user or written to disk, file system operations fail, and many times the whole system crashes.

This article by Jesse Smith on tries to prevent missleading conclusions:

It is important to note that data corruption can happen under any file system, there is nothing special about ZFS that would make it more vulnerable to corruption using non-ECC RAM.

ECC is what saves you from running into these problems, when possible, and in disastrous cases, what warns you about this happening before it's too late.