Server Fault is a question and answer site for system and network administrators. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Is it realistic to setup a 100 TB database (about 90 TB actually) on PostgreSQL without data sharding between a number of nodes? Are there any success stories / examples about similar setups?

share|improve this question

migrated from Mar 29 '11 at 6:46

This question came from our site for professional and enthusiast programmers.

I imagine it depends on your workload. How is the data distributed, and how will it be queried? What sort of response time do you require? – Frank Farmer Mar 29 '11 at 3:36
Well, the load profile may be described as frequent inserts (about 50K per second at peak), relatively seldom selects (range of rows by user and timestamp). The data may be easily sharded / partitioned by user and date/timestamp – voidlizard Mar 29 '11 at 4:41

50K writes per second that need to be absorbed is more than a challenge usually. Even in synthetic benchmarks with quite simple inserts, pgsql's limits tend to max out around roughly 10K/s - and there you don't even have such a large beast in terms of db size.

Also the IO system for that single pgsql node is going to be interesting as even with RAID10 and asuming that 50K inserts are going to be equal to just 50K IOPS (which is probably wrong - but depends on your db scheme and indices), you are going to need roughly a hundred disks paired with a very good array that saves you from buying several hundred disks to service those writes in a timely manner.

If sharding is easy and you expect such a huge write load then go for sharding. Writes can be very difficult to scale.

share|improve this answer
Agree. This is the domain of an ExaData type system. THat sad, getting 50k IOPS is quite trivial these days with SSD - otoh these are going to be expensive. I would expect a larger 7 digit budget here for the hardware, including a mid range to high end SAN. – TomTom Mar 29 '11 at 7:51
Yes, ExaData is an option if you want to go the "vertically integrated solution stack" which is probably not that bad considering the demands. – pfo Mar 29 '11 at 7:55
Yeah. There are serious advantages for something like that, an both, 100tb as well as 50.000 iops dont really scream "cheap"´. Exadata does what - 1 million IOPS when fully loaded with SSD? – TomTom Mar 29 '11 at 8:00
To add to these comments I think that given the budget required to get that volume of data with that volume of inserts I'd be tempted to use a paid-for SQL engine, it'll be a small percentage of the overall budget and you'll have much better support. – Chopper3 Mar 29 '11 at 8:51
I fully agree. The moment your budget for a SAN hits a couple of hundred thousand a lot of valuations change. – TomTom Mar 29 '11 at 11:32

I apologize for my bluntness

If you have the money or even the need for 90TB of data storage then this is exactly where you SHOULDN'T ask. There are lots of companies out there that sell this kind of product and skills ( EMC ) and for that you actually get a solid product. You are kidding yourself if you think this is an area you can save some money by doing yourself. especially if you have to ask the community.

This isn't something you try to save money on, this is something you get right the FIRST time. go call HP, Dell, EMC ect.. and ask them about this, they will be more then happy to give you PROFESSIONAL advice and also sell a product. If something like this breaks, you or the company probably wont get a 2nd chance.

A company can go bankrupt from a single broken tape backup drive, imagine the cost of 90TB of data being lost! that's not something you want to live with.

Get a professional product and if you don't think its something you can afford then i would seriously reconsider your companies priorities.

share|improve this answer
What a stupid answer. You dont even know what type of data he's storing, How about we shutdown stackoverflow and serverfault because HP, DELL and EMC exist, Why would anyone need to do it them selves when they can just pay big money to big companies ? – ADAM Sep 20 '11 at 10:30
Blunt, yes, but valid. – Andrew Wolfe Oct 15 '14 at 2:52

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.