1 Artikel

• links
  • https://www.cloudcomputing-insider.de/strategische-ansaetze-muessen-ueber-lift-and-shift-hinausgehen-a-818714/
  • https://www.heise.de/download/blog/Die-Vorteile-und-Nachteile-des-Cloud-Computing-3713041
• “Lift und shift” is one of the approaches to move things in The Cloud(tm), and it involves moving things to the cloud as-is, unchanged, without adapting them.
• Migration
  • Vorteile
    • Easy to add/reduce capacity based on demand
    • Backups etc. get easier, thing can be distributed and less likely to fail
    • Can pay for fewer onsite servers, UPS, sysadmins etc., no need to think about upgrades
  • Nachteile
    • GDPR and legal issues
    • Security issues, hackers etc
    • Accessibility:
      • You move your dependencies elsewhere:
        • Datacenters can burn and take a lot of infrastructure with them, things you don’t know you depend on
        • If your server is in your office you care less about cloudflare etc.
      • “someone else’s computer”
    • takes additional time energy skills and effort
• Cloud für mich als nutzer
  • I
    • have a server at home, previously rented a VPS
    • own my domain but outsource Gemail to Fastmail
    • Google photos, Nextcloud
    • Vorteile
      • I trust companies to be better than me at reliability
        • My cat can disconnect my home router and my server is now unreachable
        • My Vodafone connection’s SLA is worse than $rent_VPS_company_name
      • Backing up stuff is a bit easier, provider can have automated backing up solutions for your VPS
      • Need to buy fewer hardware
    • Nachteile
      • Less control when things break
      • “Distance” is harder - copying things to Latvia is longer than to an external drive
      • One more thing I forget to pay for (=Verwaltungsaufwand)

Kommunicationskosten

res = compCosts + f*(SMPLatenz*(1/N) + LANLatenz*(1-1/N)) compCosts = 1ms = 1000 mcS f = 50 SMPLatenz = 100ns = 0.1 mcS LANLatenz = 100 mcS

res = $1000 + 50*(0.1*(1/N) + 100*(1-1/N))$

>>> def fn(N):
...     res = 1000 + 50*(0.1*(1/N) + 100*(1-1/N))
...     return res

• a) Berechnen für Knoten 1-2-4-8-16-32 1: 1005.0 2: 3502.5 4: 4751.25 8: 5375.625 16: 5687.8125 32: 5843.90625
• b)
  • More nodes -> more chances it’s on another node -> more chances to need to use the 1000x longer LANLatenz instead of the SMPLatenz. After enough nodes it stops increasing as much and seems to converge at around 6000?
  • Basically mirrors slide 14:

Zeitbedarf

• a) Lesen von 10 Dateien (each 1mb) from SSD: $$10ZugriffZeit + 10(1MB/(50 MB/s)) = 101ms + 10(1/0.05 \frac{MB}{ms}) = 10+200 = 210ms$$
• b) Lesen von 1 Datei (10mb) from SSD: $ZugriffZeit + (10MB/(0.050 MB/ms)) = 1ms + 200ms = 201 ms$
• c) Übertragen von 10MB durch LAN: $1s = 1000ms$