 |
| Major hardware components and interconnection of Greenplum MR Hadoop Reference Configuration. This diagram is a screen capture of Figure 5 of this White Paper on Cisco.com. |
---
I used to say UCS C Series are designed for "smaller deployment". However, recently I found an interesting application in DCUCD v5.0 training materials, which contradicts to what I said before. Greenplum MR Hadoop Reference Configuration is a quite-large deployment with UCS C Series only. I really should correct my saying from now on!
This Reference Configuration starts from single rack: 18 sets of C210 servers. With multi-rack I believe it can be scaled up to 8 racks (144 sets of C210 servers).
Why doesn't it make use of B Series? In this design it does not rely on any external storage. It only uses "locally attached storage". Maybe this is direct result of Hadoop's distributed computing architecture: failure of any node (C series server) will not fail the whole computing system. So there is no advantages even if we use more reliable, more expensive, and centrally managed storage systems. To attach a lot of local disks, C Series would be better fit!
[Original White Paper]
Cisco and Greenplum Partner to Deliver High-Performance Hadoop Reference Configurations
---
No comments:
Post a Comment
Tip: you can also anonymously comment here.