MongoDB: 8m ops/s on 25 servers (320k per server) | Hacker News Reader

MongoDB: 8m ops/s on 25 servers (320k per server)

snailinaturtleneck.com

53 pointsdmytton17 comments

That would be very impressive if it was a real world installation. That it's an entirely artificial benchmark makes it almost entirely meaningless.

They should be bragging about the most impressive real world MongoDB setups, even if the numbers are smaller.

This was at a talk on the state of mongodb's sharding features which are still a work in progress. The demo was a preview of what's to come rather than simple bragging.

What's an "operation" in this context? Reading a boolean value? Incrementing a counter? Loading a 100kb document?

That question is spot on.

We knew already that mongo is quite fast, so why throw this useless figure into the room without any context?

"operation" is updates+inserts+reads+deletes+database commands, in ~1:1:10:1:1 ratio (IIRC).

I don't know what the exact operations were, it was supposed to simulate a real-life web app distribution of requests.

WritingToAFileDB: 300M ops/s on my netbook with an SSD.

(Where op == appending a byte.)

Not that I run a site that's anywhere close to Reddit, but I was a bit scared to hear one of their admins mention MongoDB "suffered more than one bout of catastrophic data loss in my brief testing": http://www.reddit.com/r/announcements/comments/c2spc/reddits...

320k per machine per second sounds really high, even granting high-end machines, since the non-10gen ops/s numbers I've seen have been in the 2k to 10k per second range (e.g., http://prajwal-tuladhar.net.np/2009/11/15/500/mongodbs-perfo...).

To be fair - this benchmark is from 6 months ago. There have been a lot of improvements in Mongo since then.

I also question his statement of "Slower than DB" - I have been in production w/ MongoDB since August '09 and it was a huge performance improvement over MySQL. Since then there have been incremental improvements in performance w/ each release.

Some more info: - 25 x-large ec2 instances - each instance ran mongos, mongod and java loading program - system averaged 5M "ops" per second over many hours

Average ops per server: - 3000 inserts - 25000 updates - 150000 queries - 1500 deletes - 50 getMore - 25000 command (getLastError)

Although the numbers are not possible to read from the end of the video talk[1], it looks similar to the mongostat tool[2] (albeit a web version). If that is indeed the case, then the total operations will be the sum of all inserts, queries, updates, deletes, getmores and commands per second.

Would need the 10gen guys to confirm though.

[1] http://blip.tv/file/3597247 [2] http://www.mongodb.org/display/DOCS/mongostat

Yes, exactly. The total was for all operations and, yeah, it's basically a GUI of mongostat.

Wouldn't you say the nature of the benchmark is probably the most important detail, yet it is the one that is omitted from this blog post?

Here is a link to the video from the sharding talk at MongoSF

http://blip.tv/file/3597247

i don't need 8m ops/s, but the single server durability http://jira.mongodb.org/browse/SERVER-980

Videos from MongoSF: http://www.10gen.com/event_mongosf_10apr30

were the ops sequential or random?

Reply on news.ycombinator.com