I can think of three ways to deal with lookup tables:
Store them in the database (which enforces referential integrity)
Store them in the code (which helps speed)
Store them in a db but cache them with redis or memcached.
What is your typically method when scalability is a requirement?
I've not been able to find an answer that balances reliability with scalability.
Related
I have 6 services talking to the same database SQL Server 2016 (Payments) where some services are doing write operations and some are doing read operations. Database server holds other databases as well than Payments database. We do not have any archival job in place on Payments database. We recently got 99% CPU usage and as well as memory issue on database server.
Obvious steps I can take including
Create archival jobs to migrate old data to archived database
Can scale up database server.
But still want to explore other best solutions. I have below questions.
Can we make different databases for read and write operations, if yes how?
Can we migrate data on the fly to NoSql database from RDBMS because it is faster for read operation?
What is the best design for such applications where concurrent read and write operations happens?
Storage is all about trade-offs, so it is extremely tricky to find correct "storage" solution without diving deep in different aspects such as latency, availability, concurrency, access pattern and security requirements. In this particular case, payments data is being stored which should be confidential and straightforward removes some storage solutions. In general, you should
Cache the read data, but if the same data is being modified
constantly this will not work. Caching also doesn't work well when
your reads are not public (i.e., can not be reused across multiple
read calls, preferrably across multiple users), which is possible in this case as we are dealing with payments data.
Read/write master database and read-only slaves pattern is also "common" pattern to scale reads. It doesn't scale the writes though. It again depends if the application can work with "replication lag".
Sharding is the common access pattern to scale writes. It comes with other burden of cross node query aggregation etc (in some databases).
Finally, based on the data access pattern, refactor the schema
and employ different databases. CQRS (Command Query Responsibility
Segregation) is one way to achieve it, but it comes at it has its
own pros and cons. For more details: https://learn.microsoft.com/en-us/azure/architecture/patterns/cqrs
Few years back, I read this book which helped me immensely in understanding these concepts: https://www.amazon.com/Scalability-Startup-Engineers-Artur-Ejsmont/dp/0071843655
On the slide you can see very rough architecture of booking system. It's very concurrent environment, where many users at once may try to book the same hotel/room.
At bottom we have NoSQL database, for quick response/request there is distributed cache and application which requests data.
The idea of this slide is that when you use NoSQL + Distributed Cache you'll get sync problems, means data consistency problems. You need to sync distributed cache with NoSQL db.
Question: What the solutions/techniques already exists for such case besides IMDG? That could be both frameworks or/and best practices. Is there any specific distributed caches that solves this problem?
Question2[updated]: What are the reasons we do write to the NoSQL db instead of cache? Are that transactions, node fail possibility or anything else?
P.S. That's not my slide, and author claimed that is a great use case for IMDG.
Do you really need the distributed cache? NoSQL solutions are by nature very performant, approaching the performance of stand-alone caches (like memcached).
I can get ~10ms access times out of Cassandra, which is not much slower than most caches.
I'll bet that by the time you put in cache validation overhead, and network overhead of missed cache hits, you are going to be better off going straight to your database.
You can still use caches for things that are less transient, like room types, prices, etc.
Can Vertica Database be used for OLTP data?
And if so what are the pros and cons on doing this?
Looking for a Vertica vs Oracle fight :)Since Oracle license is so costly, would Vertica do it job for a better price ?
thx all
Using Vertica as a transactional database is a bad idea. It's designed to be a data warehousing tool. Essentially, it reads and writes data in an optimized fashion. Lots of transactions? That's not what it is designed to do.
I would recommend that you look into VoltDB. Michael Stonebreaker who is the force behind Vertica founded that company as well. His basic philosophy is that Oracle, SQL Server, et al do not do well for high performance since they are designed to do everything. The future is having databases designed for specific tasks.
So he had some concepts for a data warehousing which became Vertica. For transactional databases, there's VoltDB. Not owned by HP, for the record.
For the record, I haven't used VoltDB. From what I know, it isn't as mature as Vertica is as a solution but it looks like it has a ton of promise.
HP Vertica is a column store database. The nature of the way that data is organised within a column store does not lend itself to rapid writes.
HP Vertica gets around this by having a WOS (Write Optimised Store) and ROS (Read Optimised Store which is file based).
Data is moved out of the WOS into the ROS fairly rapidly and the ROS itself has a "merge up" process that takes small ROS files and merges them together to form larger and therefore more easily scanned files.
If you tried to use Vertica for OLTP then what would happen would be that you'd get loads of ROS containers and possibly hit the default limit of 1024 ROS containers very quickly.
If you fronted the store with some form a queuing mechanism to pass through records in larger batches then this would result in fewer and larger ROS files. It would work but if you wanted to take your OLTP system to be reading very close to its writing activity it would not fit the use case.
The WOS/ROS mechanism is a neat work around for the fundamental performance penalty of writes in a column store DB but fundamentally Vertica is not an OLTP DB but rather a data mart technology that can ingest data in near real time
I think there are different ways to read into this question.
Can you use Vertica as an OLTP database?
First I'll define this question a bit. An OLTP database means the database itself is responsible for the transaction processing, not simply receiving somewhat normalized data.
My answer here is absolutely not, unless perhaps it is a single user database. There is practically no RI, no RI locking, table locks on DELETE/UPDATE, and you're likely to accumulate a delete vector in normal OLTP type usage.
You can work around some of these with some extensive middleware programming (distributed locks, heavy avoidance of DELETE/UPDATE, etc). But why? There are tons of options out there that are not Oracle, don't carry a huge price tag but give you everything you need for OLTP.
Can you use Vertica to ingest and query OLTP data?
Yes, definitely. Best to use Vertica towards its strengths, though. Queries in Vertica tend to have a fair amount of overhead, and you can plow through large amounts of data with ease, even normalized. I would not be using Vertica to primary run point queries, grabbing a few rows here and there. It isn't that you can't, but you can't with the same concurrency as other databases that are meant for this purpose.
TL;DR Use the right tool for the right job. I really love using Vertica, but just because I like to swing a hammer doesn't mean that every problem is a nail.
This question is a little old now but i'll share my experience.
I would not suggest vertica as OLTP unless you very carefully consider your workload.
As mentioned in other answers, Vertica has 2 types of storage. ROS is the Read Optimized Storage and WOS is the Write Optimized Storage. WOS is purely in memory so it performs better for inserts but queries slower as all the small updates need to be queried and unioned. Vertica can handle small loads in theory but in practice it didn't work out very well for us performance wise. Also there are drawbacks to WOS namely being that when the database fails WOS is not necessarily preserved when it rolls back to last good epoch. (ROS isn't either but in practice you lose a lot less from ROS).
ROS is a lot more reliable and gives better read performance but you will never be able to handle more than a certain number of queries without a careful design. Although vertica is horizontally scalable, in practice large tables get segmented across all nodes and therefore queries must run on all nodes. So adding more nodes doesn't mean handling more concurrent queries it just means less work per query. If your tables are small enough to be unsegmented then this might not be an issue for you.
Also worth noting is the OLTP typically implies lots concurrent transactions so you'll need to plan resource pools very carefully. By default vertica has a planned concurrency for the general resource pool of the minimum of number of cores per server or RAM/2GB. Essentially what this value does is determine the default memory allocation PER NODE for a segmented query. Therefore by default vertica will not let you run more queries than cores. You can adjust this value but once you hit a cap on memory theres no much you can do because the memory is allocated per node so adding more nodes doesn't even help. If you hit any errors at all for resource pool memory allocations that is the first config your should look at.
Additionally, Vertica is bad with deletes and updates (which resolve to a delete and an insert in the background) so if these are a regular part of your workload then Vertica is probably a bad choice. Personally we use MySQL for our dimension tables that require deletes/updates and then sync that data periodically into vertica to use for joins.
Personally I use Vertica as an OLTP-ish realtime-ish database. We batch our loads into 5 minute intervals which makes vertica happy in terms of how many/large the inserts are. These batches are inserted using COPY DIRECT so that they avoid WOS entirely (only do this if they are large batches as this forces ROS container creation and can be bad if you do it too often). As many projections as we can have are unsegmented to allow better scale out since this makes queries hit only 1 node and allocate memory on only 1 node. It has worked well for us so far and we load about 5 billion rows a day with realtime querying from our UI.
Up_one - considering the telecom use-case - are you doing CDR or something else?
To answer your original question yes Vertica may be a great fit but it depends on how you are loading the data, how you are doing updates, what your data size is and what your SLA is. I am really familiar in this space because I implemented Vertica at a telecom that I worked for at the time.
Oracle seems to have a reputation for being more scalable than other RDBMSes. After working with it a bit, I can say that it's more complex than other RDBMSes, but I haven't really seen anything that makes it more scalable than other RDBMSes. But then again, I haven't really worked on it in a whole lot of depth.
What features does Oracle have that are more scalable?
Oracle's RAC architecture is what makes it scalable where it can load balance across nodes and parallel queries can be split up and pushed to other nodes for processing.
Some of the tricks like loading blocks from another node's buffer cache instead of going to disc make performance a lot more scalable.
Also, the maintainability of RAC with rolling upgrades help make the operation of a large system more sane.
There is also a different aspect of scalability - storage scalability. ASM makes increasing the storage capacity very straightforward. A well designed ASM based solution, should scale past the 100s of terabyte size without needing to do anything very special.
Whether these make Oracle more scalable than other RDBMSs, I don't know. But I think I would feel less happy about trying to scale up a non-Oracle database.
Cursor sharing is (or was) a big advantage over the competition.
Basically, the same query plan is used for matching queries. An application will have a standard set of queries it issue (eg get the orders for this customer id). The simple way is to treat every query individually, so if you see 'SELECT * FROM ORDERS WHERE CUSTOMER_ID = :b1', you look at whether table ORDERS has an index on CUSTOMER_ID etc. As a result, you can spend as much work looking up meta data to get a query plan as actually retrieving the data. With simple keyed lookups, a query plan is easy. Complex queries with multiple tables joined on skewed columns are harder.
Oracle has a cache of query plans, and older/less used plans are aged out as new ones are required.
If you don't cache query plans, there's a limit to how smart you can make your optimizer as the more smarts you code into it, the bigger impact you have on each query processed. Caching queries means you only incur that overhead the first time you see the query.
The 'downside' is that for cursor sharing to be effective you need to use bind variables. Some programmers don't realise that and write code that doesn't get shared and then complain that Oracle isn't as fast as mySQL.
Another advantage of Oracle is the UNDO log. As a change is done, the 'old version' of the data is written to an undo log. Other database keep old versions of the record in the same place as the record. This requires VACUUM style cleanup operations or you bump into space and organisation issues. This is most relevant in databases with high update or delete activity.
Also Oracle doesn't have a central lock registry. A lock bit is stored on each individual data record. SELECT doesn't take a lock. In databases where SELECT locks, you could have multiple users reading data and locking each other or preventing updates, introducing scalability limits. Other databases would lock a record when a SELECT was done to ensure that no-one else could change that data item (so it would be consistent if the same query or transaction looked at the table again). Oracle uses UNDO for its read consistency model (ie looking up the data as it appeared at a specific point in time).
Tom Kyte's "Expert Oracle Database Architecture" from Apress does a good job of describing Oracle's architecture, with some comparisons with other rDBMSs. Worth reading.
I am going to start on with a new project. I need to deal with hundred gigs of data in a .NET application. It is very early stage now to give much detail about this project. Some overview is follows:
Lots of writes and Lots of reads on same tables, very realtime
Scaling is very important as the client insists expansion of database servers very frequently, thus, the application servers as well
Foreseeing, lots and lots of usage in terms of aggregate queries could be implemented
Each row of data may contains lots of attributes to deal with
I am suggesting/having following as a solution:
Use distributed hash table sort of persistence (not S3 but an inhouse one)
Use Hadoop/Hive likes (any replacement in .NET?) for any analytical process across the nodes
Impelement GUI in ASP.NET/Silverlight (with lots of ajaxification,wherever required)
What do you guys think? Am i making any sense here?
Are your goals performance, maintainability, improving the odds of success, being cutting edge?
Don't give up on relational databases too early. With a $100 external harddrive and sample data generator (RedGate's is good), you can simulate that kind of workload quite easily.
Simulating that workload on a non-relational and cloud database and you might be writing your own tooling.
"Foreseeing, lots and lots of usage in terms of aggregate queries could be implemented"
This is the hallmark of a data warehouse.
Here's the trick with DW processing.
Data is FLAT. Facts and Dimensions. Minimal structure, since it's mostly loaded and not updated.
To do aggregation, every query must be a simple SELECT SUM() or COUNT() FROM fact JOIN dimension GROUP BY dimension attribute. If you do this properly so that every query has this form, performance can be very, very good.
Data can be stored in flat files until you want to aggregate. You then load the data people actually intend to use and create a "datamart" from the master set of data.
Nothing is faster than simple flat files. You don't need any complexity to handle terabytes of flat files that are (as needed) loaded into RDBMS datamarts for aggregation and reporting.
Simple bulk loads of simple dimension and fact tables can be VERY fast using the RDBMS's tools.
You can trivially pre-assign all PK's and FK's using ultra-high-speed flat file processing. This makes the bulk loads all the simpler.
Get Ralph Kimball's Data Warehouse Toolkit books.
Modern databases work very well with gigabytes. It's when you get into terabytes and petabytes that RDBMSes tend to break down. If you are foreseeing that kind of load, something like HBase or Cassandra may be what the doctor ordered. If not, spend some quality time tuning your database, inserting caching layers (memached), etc.
"lots of reads and writes on the same tables, very realtime" - Is integrity important? Are some of those writes transactional? If so, stick with RDBMS.
Scaling can be tricky, but it doesn't mean you have to go with cloud computing stuff. Replication in DBMS will usually do the trick, along with web application clusters, load balancers, etc.
Give the RDBMS the responsibility to keep the integrity. And treat this project as if it were a data warehouse.
Keep everything clean, you dont need to go using a lot of third parties tools: use the RDBMS tools instead.
I mean, use all tools that the RDBMS has, and write an GUI that extract all data from the Db using well written stored procedures of a well designed physical data model (index, partitions, etc).
Teradata can handle a lot of data and is scalable.