On December 21, 2010 05:33:05 pm Peter Booth wrote: > A lot of the capacity stuff for the example you describe will be agnostic > to rails vs another toolset. Some questions: > > 1. 10M concurrently connected clients or 10M registered ?
Registered, not concurrent. > 2. 100k req / hour is 30 req/sec, which sounds trivial except that you say > each request is approx 2 MB, which means 600MBits/sec. That means you're > using most of a gig Ethernet NICs throughput. This all points to > deployment on multiple machines and all that's entailed with that. > I am expecting multiple machines, with multiple Ge or 10Ge NICs, particularly since I would be recommending Ethernet attached storage (see http://www.coraid.com/products/srx4200_3.5_sas_sata_etherdrive_storage_appliance). But I need to win the business before I start thinking about this. > Given this I'd be much more concerned about the suitability of Ruby than of > an RDBMS. You mention an average of 2MB per requests. What proportion of > your request rate do you expect will involve transferring more than say > 30k of data? Are you dynamically generating reports? > No, the reports are pregenerated. Many will have hi-res images. I need to do a lot more analysis before I can tell you what the characteristics of the request stream will look like. > > Some context: an IDE disk can do only 80 synchronous (unbufferred) IOPs. A > Fusion IO SSD card can do 80,000 IOPs. The biggest numbers I've seen > Oracle quote for write intensive workloads is about 400,000 TXN/sec on a > single server. A standard gigabit Ethernet NIC can push 950mbps with node > to node latencies of about 40usec. > Thanks for the info. Regards, Henry > Sent from my iPhone > > On Dec 21, 2010, at 1:39 PM, Owen Dall <[email protected]> wrote: > > Yes, those are definitely some hefty requirements, and if there > > requirement is to access with some kind of direct hashing algorithm that > > a NoSQL database offers would be more cost-effective... > > > > -Owen > > > > On Tue, Dec 21, 2010 at 11:44 AM, Henry Baragar > > <[email protected]> wrote: Owen, > > > > I don't have all the details yet, but it would be an Internet-based case management system that can handle: > > >10M clients > > > > 20-100 MB/client of reports and images (close to 1 PB of data) > > data for one client completely unrelated to another > > 100K requests per hour, with an average size of 2MB and a response time > > of less than 5 seconds And, these estimates could be out by a factor of > > 10 (either way). Currently they don't have these volumes, so they need a > > strategy to scale up to the anticipated volumes. It may be that a > > relational database can handle these volumes, but from what I have > > learned, from an administrative & performance tuning perspective, > > mongoDB would be much better in this situation. These requiremenst need > > to be balanced off against the development effort. Henry > > > > On December 21, 2010 07:39:57 am Owen wrote: > > > Henry, > > > > > > Can you share what write/read volume and user capacity that your client > > > thinks won't scale with a relational database? > > > > > > Perhaps that may be an issue with MySQL, but Oracle can scale (but you > > > pa $) and Postgres (free) has a great reputation, but haven't tried it > > > myself. > > > > > > -Owen > > > > -- > > Henry Baragar > > Instantiated Software -- Henry Baragar Instantiated Software -- You received this message because you are subscribed to the Google Groups "Hobo Users" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/hobousers?hl=en.
