Petabyte-scale data management
Posts about database management for databases with petabytes of user data.
Yahoo wants to do decapetabyte-scale data warehousing in Hadoop
My old client Mark Tsimelzon moved over to Yahoo after Coral8 was acquired, and I caught up with him last month. He turns out to be running development for a significant portion of Yahoo’s Hadoop effort — everything other than HDFS (Hadoop Distributed File System). Yahoo evidently plans to, within a year or so, get Hadoop to the point that it is managing 10s of petabytes of data for Yahoo, with reasonable data warehousing functionality.
Highlights of our visit included:
- There are dozens of people at Yahoo doing Hadoop development that will wind up getting open sourced. (Full-time or close to it.) In particular, everything Mark’s team does goes to open source.
- Yahoo is moving as much of its analytics to Hadoop as possible. Much of this is being moved away from Oracle and from Yahoo’s own Everest.
- A column store is being put on top of HDFS, based on Yahoo technology. Columns will be striped across nodes. Perhaps that’s why the effort is called Project Zebra.
- Mark believes that in a year Hadoop will be much further along in meeting traditional data warehousing requirements, in areas such as:
- Metadata
- SLAs/high availability/other workload management
- Data retention policies
- Security/privacy*
- Yahoo views the time-to-market benefits of Hadoop as being more important than TCO.
| Categories: Analytic technologies, Data warehousing, Hadoop, MapReduce, Open source, Oracle, Petabyte-scale data management, Web analytics, Yahoo | 6 Comments |
Facts and rumors
- Vertica is putting out a press release today touting its 100th customer, and talking of triple digit growth last year.
- Multiple sources have told me that the DATAllegro system is being thrown out of Dell, so evidently Dell is telling this to one and all. If that goes through, this would presumably leave TEOCO as DATAllegro’s single happy customer. (I haven’t checked with Microsoft for its view.)
- A rumor has it that Infiniband technology vendor Voltaire, Ltd. privately claims triple-digit sales of switches for Exadata 1 (I think that one would be one switch per Exadata installation, not per rack). Based just on a quick glance, this is far from confirmed by Voltaire’s earnings conference call transcripts or SEC filings. However, the most recent transcript does seem to indicate Voltaire got multiple Exadata deals in the telecommunications sector, and suggests some Exadata penetration in other sectors as well.
- I was told of a classified-agency user that has >1 petabyte of data on Exadata 1 and 600 terabytes or so on Netezza. My not-obviously-biased source says the agency is distinctly happier with Netezza than Exadata.
- Like ParAccel, Oracle just got dinged for TPC-related misbehavior.
- Rumor has it that Sun has no intention of helping ParAccel rerun its withdrawn TPC-H benchmark.
- ParAccel has withdrawn the claim from its home page to be the “CERTIFIED” price-performance leader. This seems to confirm that the claim was a reference to the TPC-H. In my opinion, that was a gross misrepresentation of what the TPC-H shows.
Introduction to the XLDB and SciDB projects
Before I write anything else about the overlapping efforts known as XLDB and SciDB, I probably should explain and disambiguate what they are as best I can. XLDB was organized and still is run by guys who want to solve a scientific problem in eXtremely Large DataBase Management, most especially Jacek Becla of SLAC (the organization previously known as Stanford Linear Accelerator Center). Becla’s original motivation was that he needs a DBMS to manage what will be 55 petabytes of raw image data and 100 petabytes of astronomical data total for LSST (Large Synoptic Survey Telescope). Read more
| Categories: Data models and architecture, Database diversity, Michael Stonebraker, Open source, Petabyte-scale data management, Scientific research, Theory and architecture, eBay | 1 Comment |
Facebook, Hadoop, and Hive
I few weeks ago, I posted about a conversation I had with Jeff Hammerbacher of Cloudera, in which he discussed a Hadoop-based effort at Facebook he previously directed. Subsequently, Ashish Thusoo and Joydeep Sarma of Facebook contacted me to expand upon and in a couple of instances correct what Jeff had said. They also filled me in on Hive, a data-manipulation add-on to Hadoop that they developed and subsequently open-sourced.
Updating the metrics in my Cloudera post,
- Facebook has 400 terabytes of disk managed by Hadoop/Hive, with a slightly better than 6:1 overall compression ratio. So the 2 1/2 petabytes figure for user data is reasonable.
- Facebook’s Hadoop/Hive system ingests 15 terabytes of new data per day now, not 10.
- Hadoop/Hive cycle times aren’t as fast as I thought I heard from Jeff. Ad targeting queries are the most frequent, and they’re run hourly. Dashboards are repopulated daily.
Nothing else in my Cloudera post was called out as being wrong.
In a new-to-me metric, Facebook has 610 Hadoop nodes, running in a single cluster, due to be increased to 1000 soon. Facebook thinks this is the second-largest* Hadoop installation, or else close to it. What’s more, Facebook believes it is unusual in spreading all its apps across a single huge cluster, rather than doing different kinds of work on different, smaller sub-clusters. Read more
| Categories: Data warehousing, EAI, EII, ETL, ELT, ETLT, Facebook, Hadoop, MapReduce, Parallelization, Petabyte-scale data management, Specific users, Web analytics, Yahoo | 43 Comments |
eBay’s two enormous data warehouses
A few weeks ago, I had the chance to visit eBay, meet briefly with Oliver Ratzesberger and his team, and then catch up later with Oliver for dinner. I’ve already alluded to those discussions in a couple of posts, specifically on MapReduce (which eBay doesn’t like) and the astonishingly great difference between high- and low-end disk drives (to which eBay clued me in). Now I’m finally getting around to writing about the core of what we discussed, which is two of the very largest data warehouses in the world.
Metrics on eBay’s main Teradata data warehouse include:
- >2 petabytes of user data
- 10s of 1000s of users
- Millions of queries per day
- 72 nodes
- >140 GB/sec of I/O, or 2 GB/node/sec, or maybe that’s a peak when the workload is scan-heavy
- 100s of production databases being fed in
Metrics on eBay’s Greenplum data warehouse (or, if you like, data mart) include:
- 6 1/2 petabytes of user data
- 17 trillion records
- 150 billion new records/day, which seems to suggest an ingest rate well over 50 terabytes/day
- 96 nodes
- 200 MB/node/sec of I/O (that’s the order of magnitude difference that triggered my post on disk drives)
- 4.5 petabytes of storage
- 70% compression
- A small number of concurrent users
| Categories: Analytic technologies, Data warehouse appliances, Data warehousing, Greenplum, Petabyte-scale data management, Teradata, Web analytics, eBay | 22 Comments |
Cloudera presents the MapReduce bull case
Monday was fire-drill day regarding MapReduce vs. MPP relational DBMS. The upshot was that I was quoted in Computerworld and paraphrased in GigaOm as being a little more negative on MapReduce than I really am, in line with my comment
Frankly, my views on MapReduce are more balanced than [my] weary negativity would seem to imply.
Tuesday afternoon the dial turned a couple notches more positive yet, when I talked with Michael Olson and Jeff Hammerbacher of Cloudera. Cloudera is a new company, built around the open source MapReduce implementation Hadoop. So far Cloudera gives away its Hadoop distribution, without charging for any sort of maintenance or subscription, and just gets revenue from professional services. Presumably, Cloudera plans for this business model to change down the road.
Much of our discussion revolved around Facebook, where Jeff directed a huge and diverse Hadoop effort. Apparently, Hadoop played much of the role of an enterprise data warehouse at Facebook — at least for clickstream/network data — including:
- 2 1/2 petabytes of data managed via Hadoop
- 10 terabytes/day of data ingested via Hadoop (Edit: Some of these metrics have been updated in a subsequent post about Facebook.)
- Ad targeting queries run every 15 minutes in Hadoop
- Dashboard roll-up queries run every hour in Hadoop
- Ad-hoc research/analytic Hadoop queries run whenever
- Anti-fraud analysis done in Hadoop
- Text mining (e.g., of things written on people’s “walls”) done in Hadoop
- 100s or 1000s of simultaneous Hadoop queries
- JSON-based social network analysis in Hadoop
Some Facebook data, however, was put into an Oracle RAC cluster for business intelligence. And Jeff does concede that query execution is slower in Hadoop than in a relational DBMS. Hadoop was also used to build the index for Facebook’s custom text search engine.
Jeff’s reasons for liking Hadoop over relational DBMS at Facebook included: Read more
Teradata’s Petabyte Power Players
As previously hinted, Teradata has now announced 4 of the 5 members of its “Petabyte Power Players” club. These are enterprises with 1+ petabyte of data on Teradata equipment. As is commonly the case when Teradata discusses such figures, there’s some confusion as to how they’re actually counting. But as best I can tell, Teradata is counting: Read more
| Categories: Data warehousing, Market share, Petabyte-scale data management, Specific users, Teradata, eBay | 8 Comments |
Greenplum’s single biggest customer
Greenplum offered a bit of clarification regarding the usage figures I posted last night. Everything on the list is in production, except that:
- One Greenplum customer is at 400 terabytes now, and upgrading to >1 petabyte “as we speak.”
- Greenplum’s other soon-to-be >1 petabyte customer isn’t in production yet. (Greenplum previously told me that customer was in the process of loading data right now.)
| Categories: Data warehousing, Fox and MySpace, Greenplum, Petabyte-scale data management, Specific users | 3 Comments |
Greenplum is in the big leagues
After a March, 2007 call, I didn’t talk again with Greenplum until earlier this month. That changed fast. I flew out to see Greenplum last week and spent over a day with president/co-founder Scott Yara, CTO/co-founder Luke Lonergan, marketing VP Paul Salazar, and product management/marketing director Ben Werther. Highlights – besides some really great sushi at Sakae in Burlingame – start with an eye-opening set of customer proof points, such as: Read more
| Categories: Analytic technologies, Data warehouse appliances, Data warehousing, Greenplum, Petabyte-scale data management, PostgreSQL | 18 Comments |
Yahoo scales its web analytics database to petabyte range
Information Week has an article with details on what sounds like Yahoo’s core web analytics database. Highlights include:
- The Yahoo web analytics database is over 1 petabyte. They claim it will be in the 10s of petabytes by 2009.
- The Yahoo web analytics database is based on PostgreSQL. So much for MySQL fanboys’ claims of Yahoo validation for their beloved toy … uh, let me rephrase that. The highly-regarded MySQL, although doing a great job for some demanding and impressive applications at Yahoo, evidently wasn’t selected for this one in particular. OK. That’s much better now.
- But the Yahoo web analytics database doesn’t actually use PostgreSQL’s storage engine. Rather, Yahoo wrote something custom and columnar.
- Yahoo is processing 24 billion “events” per day. The article doesn’t clarify whether these are sent straight to the analytics store, or whether there’s an intermediate storage engine. Most likely the system fills blocks in RAM and then just appends them to the single persistent store. If commodity boxes occasionally crash and lose a few megs of data — well, in this application, that’s not a big deal at all.
- Yahoo thinks commercial column stores aren’t ready yet for more than 100 terabytes of data.
- Yahoo says it got great performance advantages from a custom system by optimizing for its specific application. I don’t know exactly what that would be, but I do know that database architectures for high-volume web analytics are still in pretty bad shape. In particular, there’s no good way yet to analyze the specific, variable-length paths users take through websites.