0

Possible Duplicate:
Memcached - one large instance vs many smaller

I know this might depend on my specific scenario but if possible let's analyze it in general :)

I wonder if using 6 memcached instances of 10MB would be better than having a single instance of 60MB.

I'd say that concurrency capability would be improved but there might be a considerable performance penalty.

Community
  • 1
Claudio Redi
  • 121
  • 7
  • 2
    Instances of *what*? Tags aren't content, they're metadata. – womble Sep 19 '11 at 20:31
  • @wombie: Sorry, I copied the question from a memcached specific forum where I originally posted the question. I missed that I didn't mention memcached on the whole question. Already fixed – Claudio Redi Sep 20 '11 at 14:52

2 Answers2

1

Considering we are speaking generally...

One common reason for using multiple instances, is for improved reliability. As one server statistically could go down, but multiple down servers is much less probable.

If you are referring to using multiple instances on the same server, some benefits still do exist. There is reasons to run this type of setup, one would be data separation. You also gain some reliability due to potential process crashing issues. However, those same benefits would be better served across multiple servers.

In a completely hypothetical general situation, with absolutely no mitigating factors, the potential benefits don't justify a more complicated situation.

Mitigating factors: an unstable process, lack of concurrency, etc...; would change the question, and possibly lead to a different conclusion.

Specific to memcached, which is considered a stable and threaded process, the benefits don't justify.

J. M. Becker
  • 2,471
  • 1
  • 17
  • 21
0

If you're splitting hairs about 60 MB with an application built to handle caching gigabytes and gigabytes of data, I assure you that your bottleneck is hardware and not the memcached configuration. I can't even fathom a scenario where you have enough traffic to need six memcached instances to service your requests, but have so little data that your working set fits into that amount of RAM.

jgoldschrafe
  • 4,395
  • 18
  • 18
  • He stated that he was speaking generally, which means assuming things like a bottleneck is inappropriate for the context. – J. M. Becker Sep 23 '11 at 13:56