The IRS data warehouse
According to a recent Eric Lai Computerworld story and a 2006 Sybase.com success story,
- The IRS has a data warehouse running on Sybase IQ, with 500 named users, called the CDW (Compliance Data Warehouse). (Computerworld)
- By some metric, it’s a 150 TB warehouse. (Computerworld)
- By some metric, they add 15-20 TB/year, with a 4 hour load time. (Computerworld)
- As of 2006, there were 20-25 TB of “input data”, with a “70% compression rate”. (Sybase)
I can’t entirely reconcile those numbers, but in any case the database sounds plenty big.
Computerworld also said:
the research division also uses Microsoft Corp.’s SQL Server to store all of the metadata for the data warehouse and the rest of the agency. Managing and cleaning all of that metadata — 10,000 labels for 150 databases — is a huge task in itself,
| Categories: Analytic technologies, Data warehousing, Specific users, Sybase | 2 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.
| Categories: Analytic technologies, Columnar database management, Data warehousing, MySQL, Petabyte-scale data management, PostgreSQL, Specific users, Theory and architecture, Yahoo | 13 Comments |
Data warehouse appliance power user TEOCO
If you had to name super-high-end users of data warehouse technology, your list might start with a few retailers, credit data processors, and telcos, plus the US intelligence establishment. Well, it turns out that TEOCO runs outsourced data warehouses for several of the top US telcos, making it one of the top data warehouse technology users around.
A few weeks ago, I had a fascinating chat with John Devolites of TEOCO. Highlights included:
- TEOCO runs a >200 TB DATAllegro warehouse for a major US telco. (When we hear about a big DATAllegro telco site that’s been in production for a while, that’s surely the one they’re talking about.)
- TEOCO runs around 450 TB total of DATAllegro databases across its various customers. (When Stuart Frost blogs of >400 TB “systems,” that may be what he’s talking about.)
- TEOCO likes DATAllegro better than Netezza, although the margin is now small. This is mainly for financial reasons, specifically price-per-terabyte. When TEOCO spends its own money without customer direction as to appliance brand, it buys DATAllegro.
- TEOCO runs at least one 50 TB Netezza system — originally due to an acquisition of a Netezza user — with more coming. There also is more DATAllegro coming.
- TEOCO feels 15-30 concurrent users is the current practical limit for both DATAllegro and Netezza. That’s greater than it used to be.
- Netezza is a little faster than DATAllegro on a few esoteric queries, but the difference is not important to TEOCO’s business.
- Official price lists notwithstanding, TEOCO sees prices as being in the $10K/TB range. DATAllegro’s price advantage has shrunk greatly, as others have come down to more or less match. However, since John stated his price preference for DATAllegro as being in the present tense, I presume the price match isn’t perfect.
- Teradata was never a serious consideration, for price reasons.
- In the original POC a few years ago, the incumbent Oracle — even after extensive engineering — couldn’t get an important query down under 8 hours of running time. DATAllegro and Netezza both handled it in 2-3 minutes. Similarly, Oracle couldn’t get the load time for 100 million call detail records (CDRs) below 24 hours.
- Applications sound pretty standard for telecom: Lots of CDR processing — 550 million/day on the big DATAllegro system cited above. Pricing and fraud checking. Some data staging for legal reasons (giving the NSA what it subpoenas and no more).
| Categories: Analytic technologies, Data mart outsourcing, Data warehouse appliances, Data warehousing, DATAllegro, Netezza, Pricing, Specific users, Telecommunications, TEOCO | 7 Comments |
Outsourced data marts
Call me slow on the uptake if you like, but it’s finally dawned on me that outsourced data marts are a nontrivial segment of the analytics business. For example:
- I was just briefed by Vertica, and got the impression that data mart outsourcers may be Vertica’s #3 vertical market, after financial services and telecom. Certainly it seems like they are Vertica’s #3 market if you bundle together data mart outsourcers and more conventional OEMs.
- When Netezza started out, a bunch of its early customers were credit data-based analytics outsourcers like Acxiom.
- After nagging DATAllegro for a production reference, I finally got a good one — TEOCO. TEOCO specializes in figuring out whether inter-carrier telcom bills are correct. While there’s certainly a transactional invoice-processing aspect to this, the business seems to hinge mainly around doing calculations to figure out correct charges.
- I was talking with Pervasive about Pervasive Datarush, a beta product that lets you do super-fast analytics on data even if you never load it into a DBMS in the first place. I challenged them for use cases. One user turns out to be an insurance claims rule-checking outsourcer.
- One of Infobright’s references is a French CRM analytics outsourcer, 1024 Degres.
- 1010data has built up a client base of 50-60, including a number of financial and retail blue-chippers, with a soup-to-nuts BI/analysis/columnar database stack.
- I haven’t heard much about Verix in a while, but their niche was combining internal sales figures with external point-of-sale/prescription data to assess retail (especially pharma) microtrends.
To a first approximation, here’s what I think is going on. Read more
The eBay analytics guys have a blog now
Oliver Ratzesberger and his crew have started a blog, focusing on xldb analytics. Naturally, one of the early posts gives a quick overview of their system stats. Highlights include:
Incoming data volumes exceed 40TB per day, with more than 10^11 new items/lines/records being added per day. Our analytical processing infrastructure exceeds 6PB of physical storage with over 2.9PB(1.4+1.5) in our largest cluster.
We leverage compression technologies wherever possible and are achieving compression ratios as high as 99% on our highest volume data feeds.
On any given day our massive parallel systems process more than 27PB of data, not factoring in various levels of caches that serve similar activities or processes and reduce the amount of physical IOs significantly.
We execute millions of requests on a daily basis, spanning from near realtime highly localized access to enormous jobs that span 100s of TB in a single or series of models.
| Categories: eBay, Specific users | Leave a Comment |
More Twitter weirdness
Twitter commonly has the problem of duplicate tweets. That is, if you post a message, it shows up twice. After a little while, the dupe disappears, but if you delete the dupe manually, the original is gone too.
I presume what’s going on is that tweets are cached, the tweets are eventually batched to disk, and they don’t always get deleted from cache until some time after they’re persisted. If you happen to check the page of your recent tweets inbetween — boom, you get two hits. But what I don’t understand is why the two versions have different timestamps.
Presumably, this could be explained at a MySQL User Conference session next month, one of whose topics will be Intelligent caching strategies using a hybrid MemCache / MySQL approach. I’m so glad they don’t use stupid strategies to do this … Read more
| Categories: Cache, MySQL, OLTP, Specific users | 3 Comments |
Odd article on Sybase IQ and columnar systems
Intelligent Enterprise has an article on Sybase IQ and columnar systems that leaves me shaking my head. E.g., it ends by saying Netezza has a columnar architecture (uh, no). It also quotes an IBM exec as saying only 10-20% of what matters in a data warehouse DBMS is performance (already an odd claim), and then has him saying columnar only provides a 10% performance gain (let’s be generous and hope that’s a misquote).
Also from the article — and this part seems more credible — is:
“Sybase IQ revenues were up 70% last year,” said Richard Pledereder, VP of engineering. … Sybase now claims 1,200 Sybase IQ customers. It runs large data warehouses powered by big, multiprocessor servers. Priced at $45,000 per CPU, those IQ customers now account for a significant share of Sybase’s revenues, although the company won’t break down revenues by market segment.
| Categories: Analytic technologies, Columnar database management, Data warehousing, Pricing, Specific users, Sybase | 5 Comments |
eBay OLTP architecture
I’ve posted a couple times about eBay’s analytics side. As a companion, Don Burleson pointed me at a fascinating November, 2006 slide presentation outlining eBay’s transactional architecture and evolution. Highlights include:
- A whole lot of manual slicing of Oracle databases, so as not to exceed their capacity.
- A whole lot of careful design and ordering of transactions.
- Putting all the business logic in the application tier, with a custom O/R mapper. There’s lots of caching there, but very little state.
The presentation has a bunch of specific numbers, in case anybody wants to dive in.
| Categories: eBay, OLTP, Specific users | Leave a Comment |
Introduction to Exasol
I had a non-technical introduction today to Exasol, a data warehouse specialist that has gotten a little buzz recently for publishing TPC-H results even faster than ParAccel’s. Here are some highlights:
- Exasol was founded back in 2000.
- Exasol is a German company, with 60 employees. While I didn’t ask, the vast majority are surely German.
- Exasol has two customers. 6-8 more are Coming Real Soon. Most or all of those are in Germany, although one may be in Asia.
- Karstadt (big German retailer) has had Exasol deployed for 3 years. The other deployed customer is the German subsidiary of data provider IMS Health.
- [Redacted for confidentiality] is a strategic investor in and partner of Exasol. [Redacted for confidentiality]’s only competing partnership is with Oracle.
- Exasol’s system is more completely written from scratch than many. E.g., all they use from Linux are some drivers, and maybe a microkernel.
- Exasol runs in-memory. There doesn’t seem to be a disk-centric mode.
- Exasol’s data access methods are sort of like columnar, but not exactly. I look forward to a more technical discussion to sort that out.
- Exasol’s claimed typical compression is 5-7X. As in the Vertica story, database operations are carried out on compressed data.
- Exasol says it has performed a very fast TPC-H inhouse at the 30 terabyte level. However, its deployed sites are probably a lot smaller than that. IMS Health is cited in its literature as 145 gigabytes.
- Oracle and Microsoft are listed as Exasol partners, so there may be some kind of plug-compatibility or back-end processing story.
| Categories: Analytic technologies, Data warehousing, Exasol, Specific users | Leave a Comment |
The biggest eBay database
There’s been some confusion over my post about eBay’s multiple petabytes of data. So to clarify, let me say:
- eBay’s figure of >1.4 petabytes of data — for its largest single analytic database — counts disks or something, not raw user data.
- I previously published a strong conjecture that the database vendor in question was Teradata, which is definitely an eBay supplier. In particular, it is definitely not an Oracle data warehouse.
- While eBay isn’t saying who it is either — not even off-the-record — the 50%ish compression figures they experience just happen to map well to Teradata’s usual range.
- Edit: Just to be clear — not that there was any doubt, but I have reconfirmed that eBay is a Teradata user, in or including eBay’s Paypal division.