• nl
  • en

‘At scale, storage is the dominant hardware cost’ (iPRES2011, 6)

6 november 2011 Gepubliceerd door Laat uw gedicht achter

aaSharpeIt is not uncommon for conferences to be ‘interrupted’ by sponsor presentations. When I say ‘interrupted’, I do not necessarily mean that such talks are unwelcome. Conference days tend to be packed from early morning to late at night, and such sponsor interventions can be quite pleasant – a moment to doze off or to check your e-mail. Robert Sharpe (photo) of Tessella (vendors of the Safety Deposit Box or SDB system) gave us no such respite. In an entertaining presentation he shared some scalability experiences with us.

The case study was Family Search, which ingests no less than 20 Terabyte of images a day. That was quite a scalability test for the Tessella Safety Deposit Box system, and it tested some of Sharpe’s own assumptions:

  • Tessella expected that they would need faster, more efficient tools, but it turned out that existing tools (DROID, Jhove, etc.) were easily fast enough.
  • Tessella expected reading and writing of content to be fast compared to processing, but it turned out that reading and writing were not fast enough; the process required parallel reads and parallel writes. Thus the hardware cost is dominated by non-processing costs.
  • Tessella (and most of us) expected storage to be cheap, but at scale it turned out to be the dominant hardware cost. Reading and writing hardware came to about GBP 80,000. The storage costs came to GBP 100 per Terabyte content (3 copies), which amounted to GBP 730,000 a year, each year, and without refreshment costs.

Sharpe concluded that we do not need faster tools – but we do need better & more comprehensive tools. We need systems engineering, not just software engineering. And we need enterprise solutions: automation, multi-threading, efficient workflow management and automated issue handling.

All of which, of course, Rob will be happy to talk to you about.

PS: In response to this blog post, Rob wrote to me: ‘A further point I was trying to make in the rest of my talk is you don’t need especially powerful application servers to do this: you can do it fairly cheaply (certainly when compared to other costs at such scale).’


Scale Singapore-style: the Marine Bay Sands Hotel. The ship-like contraption on top of the three towers holds lush tropical gardens, a 150 meter swimming pool, restaurants, and a bar.

Gecategoriseerd in :Geen categorie

Geef een reactie

Het e-mailadres wordt niet gepubliceerd. Vereiste velden zijn gemarkeerd met *

Translate »
Netwerk Digitaal Erfgoed