0

For the same price we can get either

Worktsation which is 2 x 6 core Xeons 3.4GHz, 128GB RAM, 600GB SAS RAID 0

= 12 Cores, 128GB RAM

Or

Cluster 8 x (4 Core i7 3.4GHz, 16GB RAM, 250GB SATA, Intel 82579LM), Gigabit ethernet Netgear switch

= 32 Cores, 128GB RAM

Which is better? More FLOPS etc.? With the cluster having 3 times as many cores, does that make up for the gigabit ethernet interconnect? (vs the workstation with all ram on the same bus).

It's for running custom C programs doing bio-informatics.

EDIT this is a real question as the person who wants it came in asking about the two, a workstation vs a "Linux cluster" ~ £5000. From web searching I found i7 is ~80GFLOPS and Xeon is ~90GFLOPs... Workstation =~ 180GFLOPS vs 8 PCs=~ 640GFLOPS That 640GFLOPs would come down after taking into account parallelising overheads and ethernet bottleneck.

Hax
  • 139
  • 6
  • 6
    Virtually impossible to answer your question as you've provided zero information regarding the most important part - what clustering software are you looking to use and how does that work. In terms of raw cycles the 4xi7 solution is obviously more capacious but it's useless without the right clustering system, even then ethernet may have too much latency. – Chopper3 Feb 02 '12 at 16:06
  • They are looking to run a Linux cluster? Yes from the raw numbers it seems 32 cores win. But my concern is the limits from gigabit ethernet. File sizes are 2- 4GBs. – Hax Feb 02 '12 at 16:59
  • No such thing as a 'linux cluster' - there's clusters than run on linux but the servers don't just all suddenly work together magically on their own - you need some "clustering software" - hence my question. – Chopper3 Feb 02 '12 at 17:05
  • I'd imagine they would want something free and open source. Do you have any suggestions? – Hax Feb 02 '12 at 17:27
  • Ok, you're really not getting this. How about if I asked you for a recommendation for just "some form of transportation system" - could be a bike, could be an aircraft carrier or anything in between - i.e. you're telling us NOTHING here - "a cluster", to do WHAT??? – Chopper3 Feb 02 '12 at 20:48
  • 1
    Hax - you need to know what this 'cluster' is going to be actually doing, what are they wanting to use computers to actually achieve and what software do they intend to use to achieve this - CPUs alone do nothing functional, you still haven't come to use with this information. – Chopper3 Feb 03 '12 at 10:37
  • I've been trying to find out, but from what I remember the dataset is large (genome data) and the calculations are simple, as in comparisons/ finding differences. – Hax Feb 03 '12 at 10:54
  • Hax, this means NOTHING, nothing at all - are you trying to troll us, 'cos you keep coming back with zero useful information - tell whoever's asking you to go away because neither you nor them know enough to be considering this question. – Chopper3 Feb 03 '12 at 10:57

2 Answers2

2

This totally depends on the workload you need to run on the system and how well it can scale in a clustered environment.

Massimo
  • 68,714
  • 56
  • 196
  • 319
  • Agreed. For example rendering (small data, tons of calculation) is a lot different than... some sort of financial simulation (super fast file streaming for every run). It also depends on CPU usag "per core". – TomTom Feb 02 '12 at 17:27
  • Large data set, simple calculation (comparisons)? – Hax Feb 03 '12 at 10:43
  • 1
    You may as well tell us what you had for breakfast, it makes as much sense - how many times do we have to ask - tell us some real information, not just tiny peripheral rubbish. – Chopper3 Feb 03 '12 at 11:05
1

The cluster gives you a little more power and some redundancy(if it's feasible for your software). The big 'workstation' has the advantage of being simpler to deploy and won't be bottlenecked by your switch. I can't say for certain if the switch will bottleneck since it will depend on your transfer sizes, etc.

Publiccert
  • 1,110
  • 1
  • 8
  • 22