MongIOPS: Your Favorite Datastore, Only Faster

transcript

MONGIOPS Your Favorite Datastore, Only Faster…

A MONGO SERVER

This is a typical server…

December 5, 2012 Fusion-io Confidential 2

Dell R720 CPU Cores: 32 w/HT Memory: 128 GB

…is your working set > 128GB?

COST OF DRAM MODULES

4GB 8GB 16GB 32GB

Fusion-io Confidential December 5, 2012 3

$ $$ $$$

$$$$$$

A MONGO CLUSTER

With NoSQL Databases, we tend to scale out for DRAM

Combined Resources CPU Cores: 96! Memory: 384 GB

Do we need that many cores to serve reads and writes?

STORAGE IS SLOW

December 5, 2012 5 Fusion-io Confidential Source: http://www.micronblogs.com/category/history-of-digital-storage/

100,000

10,000

1988 2008

HDD Performance

TECHNOLOGY ENABLERS

FLASH MEMORY

SOFTWARE-ENABLED REPROGRAMMABLE

CONTROLLERS

PCIe Ecosystem

SSD TREATS MEMORY LIKE DISK

MONGO ON FUSION-IO

August, 2012 Fusion-io Confidential 8

Step 1: Mount ioDrive at /var/lib/mongo Step 2:

IODRIVE2 SPECIFICATIONS

ioDrive2 Capacity 400GB 600GB 365GB 785GB 1.2TB

NAND Type SLC (Single Level Cell)

SLC (Single Level Cell)

MLC (Multi Level Cell)

Read Bandwidth (1 MB) 1.4 GB/s 1.5 GB/s 910 MB/s 1.5 GB/s 1.5 GB/s

Write Bandwidth (1 MB) 1.3 GB/s 1.3 GB/s 590 MB/s 1.1 GB/s 1.3 GB/s

Read IOPS (512 Byte) 190,000 193,000 137,000 141,000 143,000

Write IOPS (512Byte) 535,000 535,000 535,000 535,000 535,000

Read Access Latency 47 µs 47 µs 68 µs 68 µs 68 µs

Write Access Latency 15 µs 15 µs 15 µs 15 µs 15 µs

Bus Interface PCI-Express 2.0 x4

Weight Less than 11 ounces

Operating Systems 64-Bit Microsoft XP/Vista/Win7/Server 2003/Server 2008, RHEL 4/5/6, SLES 10/11, OEL v5/6,

VMware ESX 4.0/4.1/ESXi 4.1/5.0*, Solaris 10 U8/U9, OSX* 10.6.7/10.7, HP-UX* 11i *ioDrive2 and ioDrive2 Duo

IODRIVE2 DUO SPECIFICATIONS

ioDrive2 Capacity 1.2TB 2.4TB

NAND Type SLC (Single Level Cell) MLC (Multi Level Cell)

Read Bandwidth (1 MB) 3.0 GB/s 3.0 GB/s

Write Bandwidth (1 MB) 2.5 GB/s 2.5 GB/s

Read IOPS (512 Byte) 385,000 285,000

Write IOPS (512Byte) 725,000 725,000

Read Access Latency 47 µs 68 µs

Write Access Latency 15 µs 15 µs

Bus Interface PCI-Express 2.0 x8

Weight Less than 16 ounces

Operating Systems 64-Bit Microsoft XP/Vista/Win7/Server 2003/Server 2008, RHEL 4/5/6, SLES 10/11, OEL v5/6,

VMware ESX 4.0/4.1/ESXi 4.1/5.0*, Solaris 10 U8/U9, OSX* 10.6.7/10.7, HP-UX* 11i *ioDrive2 and ioDrive2 Duo

GO TO OUR WEBSITE

ONE IODRIVE VS A 10-DISK RAID-0

ioDrive 2 – 1.2TB 10 x 7200 RPM SAS disks in raid-0 mdarray

YCSB DETAILS

▸  2 x 8-core 2.9GHz Xeon/64GB/2x10GbE ▸  XFS

Test ▸  Workload A: 50/50% read/write mix ▸  Workload B: 95/5% read/write mix ▸  Workload C: 100% read ▸  Workload F: 50/50% read/read+modify+write mix

Workload A - 50/50% read/write mix

Workload B - 95/5% read/write mix

Workload C - 100% read

Workload F - 50/50% read/read+modify+write mix

YCSB: LOWEST READ LATENCY WINS

December 5, 2012 18

W O R K L O A D A W O R K L O A D B W O R K L O A D F : U P D A T E W O R K L O A D F : R M W

Fusion-io Disk

Fusion-io Confidential

YCSB: LOWEST UPDATE LATENCY WINS

December 5, 2012 19

W O R K L O A D A W O R K L O A D B W O R K L O A D F : U P D A T E W O R K L O A D F : R M W

Fusion-io Disk

Fusion-io Confidential

ARCHITECTURAL IMPACT OF FUSION-IO

▸ Avoid scaling out for DRAM • Nodes can handle higher transaction load •  Terabytes of low latency persistent storage on single nodes •  Potentially avoid sharding

▸ Use less DRAM per node •  Lower cost servers • DRAM available for applications

December 5, 2012 20 Fusion-io Confidential

Talksum Data Stream Real Time for Big Data

Talksum Data Stream Appliance

•  Real Time Data Management & Analy1cs

•  Filters, Monitors, Counts, Transforms & Reduces

•  OS Level Processing for High Performance

Talksum Data Stream

System Logs

App Data

Device Stats

OtherData

Accepts Data fromany source

Filter &Route

Real-TimeMonitors & Analytics

Real Time Reduced

RemoteSites

Outputs to Storage,Applications or other Systems

BI Tools

Analytic Apps

StoredProcess

Third Party

Device

Storage

WHY DO WE USE MONGO?

▸  First Class JSON Support •  Perfect match for CEE enhanced RFC3164 •  “Schemaless” = great for log data with minor variations

▸  Capped Collections •  Control over space utilization important on an appliance

▸  Atomic Operations •  Great for working with metrics

▸  Aggregation Framework •  Convenience of SQL “group by”

▸  Map Reduce •  Familiar paradigm

ANALYZING NETFLOW DATA

Analyzing a day’s worth of Netflow Data to generate sent octets metrics for each source ip, destination ip, and source->destination ip combination using atomic increment operations

{ _id : src_key }, { $inc : { sntOcts : snt_oct } } { _id : dst_key }, { $inc : { sntOcts : snt_oct } } { _id : src_dst_key }, { $inc : { sntOcts : snt_oct }}

To test IO, messages sent as fast as possible.

No Queue! Messages dropped when IO capacity overflows.

NETFLOW MESSAGES PROCESSED A SECOND

Atomic Updates: each message is 3 Mongo operations.

RESULTS OVER 1 HOUR (3600 SECONDS)

▸ SATA Drive •  Total messages processed: 2,602,550 •  Messages Per Second: 3,614 •  Total Mongo Operations: 7,807,650 •  Mongo Operations Per Second: 10,481

▸ Fusion-io Drive •  Total messages processed: 7,811,851 •  Messages Per Second: 10,849 •  Total Mongo Operations: 23,435,553 •  Mongo Operations Per Second: 30,549

T H A N K Y O U

MongIOPS: Your Favorite Datastore, Only Faster

Technology