So are you using your EC2 instances 24/7? Do you use on-demand instances or reserved ones? EC2 seems ridiculously cheap as long as you start instances only when you need them, but for permanent usage I'm not sure yet.
- Marius On 03.08.2010, at 14:28, Simon wrote: > we run a db.m1.large instance and our db is around 20GB. previously it ran on > an intel xserve with 8GB of Ram. performance wise db.m1.large is around where > we were before, but we've not been scientific about this because we didn't > have any performance issues before, and we don't have any now. we didn't move > to RDS for performance. > > however, when staging the move we initially ran a db.m1.small instance and > that was nowhere near powerful enough. when we boot copies of production for > testing purposes we use db.m1.small, but we wouldn't use that in production. > > the real beauty of RDS with regard to performance is that is its literally a > couple of clicks to upgrade. how long do you think it will take you to > transition your DB to that a linux raid server ? we could double our compute > power and ram in literally 3 clicks and a couple of minutes - and because we > run with multi-avail zone it would automatically fail over to the slave > whilst the upgrade took place. > > we don't use ssl. traffic is limited to our ec instances, and yes sensitive > data is encrypted in the db anyway. we've just flown through PCIDSS > compliance without a glitch. > > regarding multi-avail: my understanding is that they have made limited > modifications to the 5.1 code base to support running mysql on a big scale in > the cloud. i don't know if that includes fundamental changes to the > master/slave mechanics, but the way multi-avail works "feels" like it's just > plain old replication, but wrapped in some fancy automation. > > yeah, the docs do mention latency, but we've not noticed anything at all. > > the biggest mistake we made was attempting to run apps outside ec2 pointing > at RDS. the latency in that set-up killed our apps. ymmv. > > simon > > > > On 2 August 2010 21:02, Kieran Kelleher <kieran_li...@mac.com> wrote: > Sounds great Simon. > > I have a database of about 35GB of data running on an 8GB PowerPC G5 today in > one of my active projects and we have preliminary plans under way to upgrade > our DB server to a 32GB Linux RAID unit. What is the biggest RDS memory size > instance that you have used, and what is the perception of performance gains, > if any, over traditional self or colo hosting? > > I notice they support SSL connections also to MySQL. Do you use SSL between > the EC2 app instance and the RDS instance - or is that overkill considering > that I have sensitive data (credit card numbers, etc) encrypted in the > database fields anyway? If you do use SSL connections between app and db, > have you noticed much latency? > > You said you have availed of the different zone replication/failover feature > - from reading the FAQs, it appears that this is different to traditional > master-slave replication - are they executing the SQL in parallel on both the > master and failover RDS instances to give true mirroring, or am I reading > this wrong? Have you noticed latency impact due to this configuration (the > online info suggests that there is some latency)? > > Regards, Kieran > > On Aug 2, 2010, at 1:38 PM, Simon wrote: > >> How does session management work with the elastic load balancer? For example >> if you have 3 independent EC2 instances all running the same app? >> >> if you are not using https then amazon provide a couple of cookie-based >> mechanisms for session stickiness. if you are using https then you can use >> the elb to send initial requests to one of your instances, then the user >> communicates with that specific instance directly. there is no >> ssl-termination available with elb, but the amazon lists suggest this is >> coming. once they have this ssl load balancing will be a lot more elegant. >> >> Also, do you completely trust RDS to make sure your data is never lost? Is >> there any need for you to have a physical server replicating from RDS? Is >> there any risk that one day, amazon loses your database and says "Sorry, but >> we assume you have your own backup"? >> >> in short, yes, i completely trust it. we've been running it in production >> for 9 months now without a single glitch. we use their multi-avail support >> and we've done test failovers which happen flawlessly in minutes. how long >> would it take you to (a) make a decision to fail over your master to a slave >> and (b) physically carry out the failover and (c) physically restore the >> master once things are sorted out ? the automation here alone makes it a >> much more powerful solution than running it ourselves. >> >> and how often do you test restoring from your backups ? officially we used >> to do it once a month, but it was always a real drag... now we routinely >> restore databases - sometimes several times a day - and use them to test >> code against because it's 2 clicks, make a cup of tea, and you've got a >> fully functioning snapshot of production from 5 minutes ago. >> >> do we ever take "normal" backups ? yes, but very very rarely, and not for >> date protection - we do them purely to get a fresher copy on our laptops for >> offline use. >> >> Simon >> >> >> >> -Kieran >> >> On Jul 27, 2010, at 1:16 PM, Simon wrote: >> >>> doing what you've done means you're managing mysql, looking after it, >>> making sure it doesn't fall over, doing backups, managing replication etc. >>> rds does all of that for you. it also makes changing the config of your >>> database server a breeze: need more disk space ? couple of clicks. need >>> more ram ? couple of clicks. need more compute power behind it ? couple of >>> clicks. need automatic fail-over to a different availability zone ? couple >>> of clicks. >>> >>> re web server resources, remember it's just a normal wo deployment running >>> in the cloud, so you can do whatever you do now. >>> >>> we don't separate the web and app tier - all our ec2 instances run monitor, >>> wotaskd and apache, and are effectively independent of each other, and we >>> use an elastic load balancer up front. >>> >>> simon >>> >>> >>> On 27 July 2010 17:40, James Cicenia <ja...@jimijon.com> wrote: >>> So the base image is the actual OS? So you are managing it as the admin? >>> >>> I decided to try WOlastic. I configured the instances, setup up mysql with >>> my users and sync'd the database from existing production to amazon. >>> So you are suggesting RDS vs. what I just did? What are the benefits of >>> RDS? Amazon backs up the mysql I created. >>> >>> Now I am a bit stumped on WebServerResources. How are you handling that? >>> >>> Well, if this works well, I can my webobject apps over and then just sell >>> my server and drop the colo. >>> >>> - James >>> >>> On Jul 27, 2010, at 11:28 AM, Simon wrote: >>> >>>> rolling your own is surprisingly easy if you start with a base image. we >>>> started out with a vanilla centos image from rightscale, and have built it >>>> up into what we needed from there. you can then create an ebs-backed ami >>>> in a couple of clicks. >>>> >>>> re pricing, it all depends on what you need. our financial models tell us >>>> for our deployment is excellent value for money, and we can scale well >>>> beyond our current needs and it remains as such. use the cost aws >>>> calculator to figure out your own costs, and remember to factor in staff >>>> costs in your decision making process. those DBA's are darn expensive >>>> compared to RDS :-) >>>> >>>> http://calculator.s3.amazonaws.com/calc5.html >>>> >>>> the only performance issue we found is that it is basically impossible to >>>> host your DB outside of amazon due to latency. but you don't have to use >>>> RDS - if you like sticking needles in your eyes you can just run and look >>>> after your own mysql / postgre / mssql / whatever on an ec2 instance. >>>> >>>> the general performance of our apps has also vastly improved. a mixture of >>>> using more computing power and amazon having much faster internet transit >>>> than we were paying for in our previous co-lo. >>>> >>>> alongside production we also run our staging servers and our hudson build >>>> server on ec2. in productivity terms running hudson there was a huge leap >>>> forward: previously a new build would take around 30 minutes to upload to >>>> staging / production. now it takes 19 seconds flat :-) >>>> >>>> we're shortly going to move our subversion repository to ec2 as well. >>>> >>>> Simon >>>> >>>> On 27 July 2010 15:13, James Cicenia <ja...@jimijon.com> wrote: >>>> This is very cool. >>>> >>>> I need to move one of my servers, or, use the cloud approach for its >>>> WOApps. I see you rolled your own but wolastic seems like it is for a mere >>>> mortal. >>>> >>>> Anyone use wolastic? What is the pricing your are seeing? Issues? >>>> Performances? Etc. >>>> >>>> Thanks. >>>> James Cicenia >>>> >>>> >>>> >>>> On Jul 26, 2010, at 3:55 PM, Simon wrote: >>>> >>>>> we don't use the wolastic images (we have our own) but we do deploy >>>>> entirely on the amazon ec2 cloud now. ec2 instances running standard >>>>> javamonitor / wotaskd, amazon RDS for database server, s3 for file >>>>> storage etc. scalability on demand, load balancing, redundancy across >>>>> multiple availability zones. it's the best thing since sliced bread... >>>>> >>>>> our staging servers (also on ec2) run wonders javamonitor / wotasd and >>>>> hence we'll probably upgrade our production servers to those soon. >>>>> >>>>> simon >>>>> >>>>> On 26 July 2010 21:36, Ramsey Gurley <ram...@xeotech.com> wrote: >>>>> I haven't tried it yet, but WOlastic looks like a *really* cool >>>>> deployment solution for WO. >>>>> >>>>> http://wolastic.com/ >>>>> >>>>> Ramsey >>>>> >>>>> >>>>> On Jul 26, 2010, at 4:27 PM, Ken Anderson wrote: >>>>> >>>>> Thanks for the thoughts guys! >>>>> >>>>> Ken >>>>> >>>>> On Jul 26, 2010, at 1:42 PM, Pascal Robert wrote: >>>>> >>>>> >>>>> Le 2010-07-26 à 12:55, Chuck Hill a écrit : >>>>> >>>>> On Jul 26, 2010, at 9:44 AM, Ken Anderson wrote: >>>>> >>>>> I've been asked to comment on the best way to deploy WebObjects today >>>>> without any "imposed" restrictions. I haven't done any new deployments >>>>> in a long while, so I'm likely not up to date on the last. What are >>>>> people using today, and why do they think it's the best? >>>>> >>>>> Thanks much! >>>>> Ken >>>>> >>>>> Lacking imposed restrictions (e.g. must run in J2EE container), >>>>> traditional WO deployment through Apache with mod_webobjects is probably >>>>> the way to go. Anjo was working on mod_proxy deployment, but I don't >>>>> recall how far he got or if he has this in production. It looked >>>>> promising. There is also a Fast CGI adaptor and Ravi is working on >>>>> something for WOWODC. >>>>> >>>>> I'm adding some mods in JavaMonitor too (for WOWODC) and Andrew Lindesay >>>>> also have stuff in LEWOStuff to use mod_proxy_ajp. >>>>> >>>>> >>>>> >>>>> ---- >>>>> Pascal Robert >>>>> prob...@macti.ca >>>>> >>>>> AIM: MacTICanada >>>>> Twitter : MacTICanada >>>>> LinkedIn : http://www.linkedin.com/in/macti >>>>> WO Community profile : http://wocommunity.org/page/member?name=probert >>>>> >>>>> >>>>> _______________________________________________ >>>>> Do not post admin requests to the list. They will be ignored. >>>>> Webobjects-dev mailing list (Webobjects-dev@lists.apple.com) >>>>> Help/Unsubscribe/Update your Subscription: >>>>> http://lists.apple.com/mailman/options/webobjects-dev/ramsey%40xeotech.com >>>>> >>>>> This email sent to ram...@xeotech.com >>>>> >>>>> >>>>> _______________________________________________ >>>>> Do not post admin requests to the list. They will be ignored. >>>>> Webobjects-dev mailing list (Webobjects-dev@lists.apple.com) >>>>> Help/Unsubscribe/Update your Subscription: >>>>> http://lists.apple.com/mailman/options/webobjects-dev/simon%40potwells.co.uk >>>>> >>>>> This email sent to si...@potwells.co.uk >>>>> >>>>> _______________________________________________ >>>>> Do not post admin requests to the list. They will be ignored. >>>>> Webobjects-dev mailing list (Webobjects-dev@lists.apple.com) >>>>> Help/Unsubscribe/Update your Subscription: >>>>> http://lists.apple.com/mailman/options/webobjects-dev/james%40jimijon.com >>>>> >>>>> This email sent to ja...@jimijon.com >>>> >>>> >>> >>> _______________________________________________ >>> Do not post admin requests to the list. They will be ignored. >>> Webobjects-dev mailing list (Webobjects-dev@lists.apple.com) >>> Help/Unsubscribe/Update your Subscription: >>> http://lists.apple.com/mailman/options/webobjects-dev/kieran_lists%40mac.com >>> >>> This email sent to kieran_li...@mac.com >> >> > > > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Webobjects-dev mailing list (Webobjects-dev@lists.apple.com) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/webobjects-dev/m.soutier%40starhealthcare.info > > This email sent to m.sout...@starhealthcare.info
_______________________________________________ Do not post admin requests to the list. They will be ignored. Webobjects-dev mailing list (Webobjects-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/webobjects-dev/archive%40mail-archive.com This email sent to arch...@mail-archive.com
