I'm a Newb looking for a little advice
26 views
Skip to first unread message
G W
Nov 24, 2015, 10:38:06 AM11/24/15
to SolrNet
A little background. I come from the iron age and have a little trouble believing I can deploy on Google cloud effectively.
I was a developer when Microsoft did not support network cards. everything was 3rd party and Trumpet Winsock was just a passing fad. With decades of RDBMS behind me I find Solr/Hadoop to be a bit of a nightmare. So far I'm loving it because it is new to me.
This is actually the first question post to a forum I've made in my entire career. I've responded to a few posts, read posts but never actually made a post. I hope I'm in the right place.
My plan is to deploy a SolrCloud with 5 or more pieces of iron.
My problem is, there seems to be no absolute rules and everything seems like some kind of black art when planning a system. I can't merely tap a calculator.
My primary document core has has a schema consisting of 11 fields. All are small text except one field which is about 100 words max. In terms of RDBMS everything would be a flat file. Currently I am using Solr as the DB. In my opinion the record is small. The record set or doc collection will be over 100M so the collection is going to be huge.
It seams ludicrous to store 11 field records in yet another system when everything is flat file.
My main problem is, my project which has an NDA attached has the possibility of going viral. All I seem to have is the knowledge that Solr/Hbase can do it.
I plan to set up three basic Zookeepers on dedicated hardware. These machines will be vertically scalable to insanity.
Then I plan to set up a Solrcloud of 5 servers running Intel SSD in RAID 0. From my reading, Intel SSD is the only drive that supports drive caches in a RAID. 1 boot, 6 in RAID 0, one hot standby is the current plan. The Solr boxes will be a single XEON quad with 32G Ram.
So the question is? Am I crazy? I can't really see why a jbod would be better than Raid 0 in Solr. I can sort of see why a bunch of volumes can make sense in hbase.
Anyone want to help an old dog with some new tricks?
Thanks in advance,
GW
I was a developer when Microsoft did not support network cards. everything was 3rd party and Trumpet Winsock was just a passing fad. With decades of RDBMS behind me I find Solr/Hadoop to be a bit of a nightmare. So far I'm loving it because it is new to me.
This is actually the first question post to a forum I've made in my entire career. I've responded to a few posts, read posts but never actually made a post. I hope I'm in the right place.
My plan is to deploy a SolrCloud with 5 or more pieces of iron.
My problem is, there seems to be no absolute rules and everything seems like some kind of black art when planning a system. I can't merely tap a calculator.
My primary document core has has a schema consisting of 11 fields. All are small text except one field which is about 100 words max. In terms of RDBMS everything would be a flat file. Currently I am using Solr as the DB. In my opinion the record is small. The record set or doc collection will be over 100M so the collection is going to be huge.
It seams ludicrous to store 11 field records in yet another system when everything is flat file.
My main problem is, my project which has an NDA attached has the possibility of going viral. All I seem to have is the knowledge that Solr/Hbase can do it.
I plan to set up three basic Zookeepers on dedicated hardware. These machines will be vertically scalable to insanity.
Then I plan to set up a Solrcloud of 5 servers running Intel SSD in RAID 0. From my reading, Intel SSD is the only drive that supports drive caches in a RAID. 1 boot, 6 in RAID 0, one hot standby is the current plan. The Solr boxes will be a single XEON quad with 32G Ram.
So the question is? Am I crazy? I can't really see why a jbod would be better than Raid 0 in Solr. I can sort of see why a bunch of volumes can make sense in hbase.
Anyone want to help an old dog with some new tricks?
Thanks in advance,
GW
Reply all
Reply to author
Forward
0 new messages