| Grid computing has been around for a few years | | | | include Platform LSF, Sun Grid Engine, Data Synapse, |
| now and its advantages are many. Grid computing | | | | PBS, Condor, UnivaUD, among others. Each do a |
| can be defined in many ways but for these | | | | good job of monitoring resources and managing job |
| discussions let's simply call it a way to execute | | | | submissions based upon internal policy engines. |
| compute jobs (e.g. perl scripts, database queries, etc.) | | | | 5) This model scales very well. Need more compute |
| across a distributed set of resources instead of one | | | | resources just plug them in by installing grid client on |
| central resource. In the past most computing was | | | | additional desktops or servers. They can be removed |
| done in silos or large SMP like boxes. Even today you'll | | | | just as easily on the fly. |
| still see companies perform calculations on large SMP | | | | 6) Upgrading can be done on the fly without |
| boxes (e.g. E10K's, HP Superdomes). But this model | | | | scheduling downtime. Since there are so many |
| can be quite expensive and doesn't scale well. | | | | resources some can be taken offline while leaving |
| Along comes grid computing and now we have the | | | | enough for work to continue. This way upgrades can |
| ability to distribute jobs to many smaller server | | | | be cascaded as to not effect ongoing projects. |
| components using load sharing software that | | | | 7) Jobs can be executed in parallel speeding |
| distributes the load evenly based on resources and | | | | performance. Using things like MPI will allow message |
| policies. Now instead of having one heavily burdened | | | | passing to occur among compute resources. |
| server the load is spread evenly across many smaller | | | | Some disadvatages: |
| computer which can be spread around various | | | | 1) For memory hungry applications that can't take |
| locations. | | | | advantage of MPI you may be forced to run on a |
| Some advantages are quite obvious. | | | | large SMP |
| 1) No need to buy large SMP servers for applications | | | | 2) You may need to have a fast interconnect |
| that can be split up and farmed out to smaller | | | | between compute resources (gigabit ethernet at a |
| servers (which cost far less than SMP servers). | | | | minimum). Infiband for MPI intense applications |
| Results can then be concatenated and analyzed upon | | | | 3) Some applications may need to be tweaked to |
| job(s) completion. | | | | take full advantage of the new model. |
| 2) Much more efficient use of idle resources. Jobs | | | | 4) Licensing across many servers may make it |
| can be farmed out to idle server or even idle | | | | prohibitive for some apps. Vendors are starting to be |
| desktops. Many of these resources sit idle especially | | | | more flexible with environment like this. |
| during off business hours. | | | | Areas that already are taking good advantage of grid |
| 3) Grid environments are much more modular and | | | | computing include bioinformatics, cheminformatics, oil |
| don't have single points of failure. If one of the | | | | & drilling, and financial applications. |
| servers/desktops within the grid fail there are plenty | | | | With the advantages listed above you'll start to see |
| of other resources able to pick the load. Jobs can | | | | much larger adoption of Grids which should benefit |
| automatically restart if a failure occurs. | | | | everyone involved. I believe the biggest barrier right |
| 4) Policies can be managed by the grid software. | | | | now is education. |
| Some of the most popular grid enabling software | | | | |