Throw More Boxes At It?

As commodity hardware became cheaper, a common approach to scaling up has been to just throw more boxes at it. Individual efficiency (hardware or software) and reliability doesn't really matter too much and it is not worth the time to optimize the setup when it is cheaper to order a few hundred more machines whenever a bit more capacity is needed. Repeat as needed.

More recently this trend has started to hit some limits. The individual boxes are cheap but acquiring thousands is still an expense. More importantly, the cost of electricity to run them all and to run the A/C to cool the buildings is rapidly escalating. The real estate footprint costs are also a concern.

None of this is news if you follow the trade press even casually, of course. Articles about the concern over electrical and space costs have been all over for the last few years.

We've seen the hardware adapt, with CPUs which deliver more throughput per watt. The most exciting story on this front is certainly the Sun Niagara server line. If you haven't already, go try and buy one and see how good it can be. Practices are also changing, as seen from the renewed interest in combining a number of under-utilized servers into one hardware box (see virtualization).

Fewer boxes, less electricity, less A/C, less space, fewer purchase orders, less administration. What's not to like?

Curiously, there doesn't (yet??) seem to be quite the same buzz on the software side on this issue. It's cool to reduce machine (and watt) counts by upgrading to efficient hardware like the T2000 but how about even further reducing the boxes needed by upgrading to more scalable software?

I was wondering about this while reading the SAMP announcement. Why pay for higher Watts/HTTP-request running Apache if you could run JES Web Server 7.0 instead (unlike the T2000 try & buy, the web server is try and ... deploy, being free for production as well).

How do you feel about it? Is software efficiency & scalability going to become increasingly important as the electrical costs of running massive server farms continues to increase?

Comments:

I think this is an area that can be exploited even more: Apache + Generic Hardware == OK Apache + T2000 == Good Web Server + Generic Hardware == Good Web Server + T2000 == Great By combining both more efficient hardware with more efficient software total watts consumed, total real estate consumed, total administration costs, and mroe all come down. How neat is that?

Posted by Joe McCabe on February 26, 2007 at 06:18 AM PST #

Indeed. It would be interesting to see comparative tests of those and other deployment scenarios. I know of the keylabs report (http://keylabs.com/results/sun/SunONEFinalReport_Solaris.pdf) but that's several years old now and performance has improved since then. Anyone done any (independent) tests lately?

Posted by Jyri on February 26, 2007 at 08:47 AM PST #

Post a Comment:
Comments are closed for this entry.
About

jyri

Search

Top Tags
Categories
Archives
« April 2014
SunMonTueWedThuFriSat
  
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
   
       
Today