Greybeards talk car videos, storage and IT trends with Marc Farley

In our 30th episode, we talk with 3rd time guest star,  Marc Farley (@GoFarley), Formerly with Datera and Tegile. Marc has recently gone on sabbatical and we wanted to talk to him about what was keeping him busy and what was going on in storage/IT industry these days.

Marc is currently curating a car comedy vlog called theridecast.com. Apparently people, at least in California, are making comedy videos in their cars. They can be quite hilarious, checkout this episode of comedian in cars getting coffee.

While in the storage biz, the industry is getting battered by a number of trends: IT shrinking budgets, vendor proliferation, migration to cloud, and flash becoming old hat. Marc makes multiple points as to why the storage market is undergoing such a major transition these days:

  • Death to tech refresh, long live the cloud –  yes the cloud does upgrade hardware but  planned storage system obsolesce doesn’t happen in the cloud anymore. Cloud providers are  buying new SSDs, disks, white box servers, memory etc,  but not enterprise class storage, server or networking hardware.
  • AFA is boring, but selling – every vendor’s got one , two or sometimes three and they all know how to provide flash storage services. Customers pay extra for AFA, whether they need to or not, because they are swapping out old expensive, enterprise class storage for AFAs that often cost less but still provide better performance..
  • Tail IO latency becoming more important but it’s not understood – when IO response times go from 100µsec to 10msec, it hurts. It doesn’t matter if it’s every 1000 or 10,000 IOs, customers want less performance variability, which is a main reason they move to AFA in the first place. But not all AFA’s perform the same in tail latency and SSD controller/system architecture make a big difference.
  • Hybrid storage survives but only if you go big – hybrid storage economics makes sense only for large, diverse data repositories, that mix user directories, non-performance sensitive apps, and other structured and unstructured data in one data store.
  • Greenfield apps & secondary storage are moving to the cloud but migrating current apps to the cloud is difficult –  for new app development and archive storage, moving to or starting in the cloud is a no-brainer. Transitioning running enterprise class apps to the cloud is tough to do, that requires multiple skill sets and may never be successful. Hybrid  (cloud-on premises) enterprise class apps are too arduous to even contemplate.
  • Realtime analytics is emerging but data needs to be on flash – yes MapReduce is a batch activity which can uses lots of slow disk but there’s more to analytics than MR, and doing log analysis, in anything approaching realtime, one needs flash performance.
  • Optical’s persistence is great but who leaves data on the same technology for  20 years –  with magnetic and electronic storage densities going up every couple of years, who could afford keep data on the same optical technology that was 20 years old. Imagine using microfiche to keep PB of data today, inconceivable.

As for IT in general, one limiter of IT activity will become the lack of skilled engineers, specifically full-stack engineers and data scientists.

We ended our discussions on the economics of Samsung 3D NAND and Intel-Micron (IM) 3D Xpoint non-volatile memories. Both new semiconductor technologies are always long term investments. Today, Samsung is probably losing money on each 3D TLC NAND SSD it sells, but over time, as  fab yields improve, it should become cheap enough to make a profit. Similarly, 3D Xpoint may be costly to produce early on, but as IM perfect  their fab processes, the technology should become inexpensive enough to make oodles of $s for them. And there’s more technology changes to come.

The podcast runs just over 40 minutes and covers a lot of ground. Marc’s been in the IT almost as long as the GreyBeards and has a unique perspective on what’s happening today, having been with so many diverse, major and (minor) startup vendors throughout his tenure in the industry.  Listen to the podcast to learn more.

Marc Farley


Marc is a storage greybeard who has worked for many storage companies and is currently on sabbatical. He has written three books on storage including his most recent, Rethinking Enterprise Storage: A Hybrid Cloud Model and his previous books Building Storage Networks and Storage Networking Fundamentals.

In addition to his writing books he has been a blogger and podcaster about storage topics while working for EqualLogic, Dell, 3PAR, HP, StorSimple,  Microsoft, and others.

When he is not working, Marc likes to ride bicycles, listen to music, spend time with his family and dote on his cats. Of course there’s that car video curation…

2 thoughts on “Greybeards talk car videos, storage and IT trends with Marc Farley

  1. This was a tremendous podcast. I particularly enjoyed the discussion on tail latency. I think this is a topic that would be perfect for a focused future podcast, since I agree with you that it is both important and misunderstood.

    1. Jeff, thanks for your comment. Tail latency is a very misunderstood phenomenon and exists in disk-only, all-flash and hybrid storage systems, although the magnitude of each system’s tail latency may be significantly different.

      The reasons for tail latency have to do with the drive underlying technology, its controllers, error handling, the storage system it’s attached to and the amount and handling of any caching done for the storage at each and every layer in the file-storage-system stack. Although the reasons for its existence are relatively well known, the ability to eliminate it or rather to shrink the delta between a storage systems average/median IO latency and its tail (or 99.9%) latency is not settled science as far as I can see. Yes, all flash can take the orders of magnitude down from Msec to 100s of Usecs but it never goes away.

      It continues to haunt every storage system I have ever seen/worked on. But it’s a silent phenomenon. Nobody (prior to Google’s research paper) talks about it much. Most benchmarks never report on it, although for some in the detailed reports you can gather some information about it (See SPC-1 reports). I am as guilty as most, I seldom talk about it in my performance reporting (see my monthly newsletter) and it’s only recently that I even started to gather the data that was buried in the details of some benchmarks.

      But it’s starting to get some press. Maybe some more benchmarks will start reporting on it, maybe some enterprising individuals will start reporting on it.

      As for making it an idea for a podcast, we will take it under advisement, however I think a better approach might be to talk about it in a blog post/performance report. Howard and I don’t go into any podcast with any preconceived ideas about what we plan to discuss (maybe you might have noticed). The ideas come in real time wherever the conversation leads us. Some vendors find this disconcerting, but Howard and I wouldn’t have it any other way. -Ray

Leave a Reply