From: Andrea A. <and...@ge...> - 2011-02-02 21:20:13
|
On Mon, Jan 31, 2011 at 9:12 AM, Arne Kepp <ar...@ti...> wrote: > Hi, > > lots of good points, but to me it reads like you are really confident > that your shared storage backend will not fail. I don't know whether > that's based on a big SAN or NAS. In the organizations I've worked for > in the past two years, neither of those types of solutions have > impressed with their performance or reliability (and especially not with > the reliability of their performance). For many GeoServer / GeoWebCache > users these solutions are out of question because of the price, and it's > not an option in the cloud either. Err... I guess I don't understand what you mean by failure. If the SAN/NAS fails as in starting losing data we're screwed no matter what mechanism is used no? The weak point of the temp file would be not having the atomic rename support: are SAN not going to support it properly, and if so, what would be the expected outcome? > One thing you may have overlooked is that the metastore (when clustered) > provides the locking mechanism you're recreating. It writes a timestamp > to the appropriate row, which works like the temporary file you want to > use. H2 hasn't evolved as quickly as I thought it would, and it's time > to look elsewhere. But if you do the temporary file thing you must also > add something to prevent staleness, on SANs that can be tricky (copy on > write). We don't have much of a staleness problem in our case as the data is going to be replaced all at once (no many little changes, one full layer rewrite every n months instead). About using the metastore... if I use a database that kind of environment has only and only one acceptable answer: Oracle. Would the metastore be usable against an Oracle db? A bit worried about the overhead though, using it would require a sql query against a db sitting on another box for each and every tile access, right? > Like Gabriel, I think distributed storage, such as Cassandra, would be > the biggest step forward for GWC. So I would be really interested in > talking to someone who understands Cassandra or a nosql system that is > more suitable for storing tiles + a few extra columns. I'm not sure > locking is that important, as you say, it's not a big deal if we > duplicate some data. A good backend will do that anyway, to avoid > storage hot spots. Hmmm... that may be a though sell... or else, something to swipe under the carpet very carefully, at least in the environment I'm thinking about. Cheers Andrea -- Ing. Andrea Aime Technical Lead GeoSolutions S.A.S. Via Poggio alle Viti 1187 55054 Massarosa (LU) Italy phone: +39 0584962313 fax: +39 0584962313 http://www.geo-solutions.it http://geo-solutions.blogspot.com/ http://www.linkedin.com/in/andreaaime http://twitter.com/geowolf ----------------------------------------------------- |