| 1 |
More info: www.blkreplay.org |
| 2 |
|
| 3 |
---------------------------------------------------------------- |
| 4 |
|
| 5 |
Abstract: |
| 6 |
|
| 7 |
blkreplay is a GPL’ed toolkit, driving the block |
| 8 |
layer of Linux (or other Unix-like OSes) while measuring |
| 9 |
latency and throughput of IO operations for later visualization |
| 10 |
(so-called “sonar diagrams” and others). |
| 11 |
|
| 12 |
In addition to artificial loads like random read-write sweeps |
| 13 |
and various kinds of overload tests, blkreplay can replay |
| 14 |
**natural loads** which have been recorded by blktrace (or |
| 15 |
Windows DiskMon) at heavily-loaded production servers |
| 16 |
at big data centers. |
| 17 |
|
| 18 |
blkreplay comes with a **large collection of natural loads** |
| 19 |
from a wide spectrum of applications (such as web servers, |
| 20 |
databases, dedicated servers, etc) which have been released |
| 21 |
to the public by 1&1 under GPL. |
| 22 |
|
| 23 |
See http://www.blkreplay.org/loads/ |
| 24 |
|
| 25 |
Some of these natural loads have recorded the real-life |
| 26 |
disk access behaviour from servers serving thousands of |
| 27 |
customers in parallel. |
| 28 |
|
| 29 |
Static analysis of the workingset behaviour of such natural |
| 30 |
loads is also implemented. |
| 31 |
|
| 32 |
blkreplay comes with a modular and extensible **test suite**, |
| 33 |
automating large projects for testing and/or benchmarking. |
| 34 |
|
| 35 |
blkreplay can be used to test physical hardware, e.g. compare |
| 36 |
different brands of hard disks, or RAID controllers / their |
| 37 |
settings / RAID rebuild performance degradation, or to |
| 38 |
evaluate the effect of SSD caching, or to compare different |
| 39 |
block level transports like iSCSI vs Fibrechannel (over |
| 40 |
different kinds of storage networks). |
| 41 |
|
| 42 |
It can compare virtual hardware (like vmware or XenServer |
| 43 |
block devices, or any type of block-level storage |
| 44 |
virtualization) to each other or to physical hardware, |
| 45 |
provided the test setup is handled very carefully. |
| 46 |
|
| 47 |
blkreplay can compare commercial storage boxes from vendors |
| 48 |
like EMC, NetApp, IBM, Hitachi, etc to each other or to |
| 49 |
cheap off-the-shelf hardware (in order to determine the |
| 50 |
price/performance ratio), provided the test setup is also |
| 51 |
handled very carefully. |
| 52 |
|
| 53 |
Furthermore, it can be used for tests of the Linux kernel, |
| 54 |
e.g. for testing device drivers, comparing different IO |
| 55 |
schedulers at different load patterns, determining the |
| 56 |
overhead of Linux dm targets, determining the impact of |
| 57 |
network problems to DRBD, and much more. |
| 58 |
|
| 59 |
At 1&1, blkreplay has even been used as a tool for |
| 60 |
root cause analysis of incidents: for example, high |
| 61 |
load peaks presumably stemming from traffic jam (or other |
| 62 |
sources of overload) were recorded at production |
| 63 |
sites in real time by blktrace, and later replayed |
| 64 |
in the laboratory (without causing customer impact) |
| 65 |
seeking for the cause of trouble, or improving the |
| 66 |
safety margins by choice of better hardware. |
| 67 |
|
| 68 |
For experts in IO subsystems, visualization techniques like |
| 69 |
“sonar diagrams” can reveal (parts of) the internal structure |
| 70 |
of complex IO systems, such as cache hierarchies or other |
| 71 |
hierarchical storage systems. |
| 72 |
|
| 73 |
As a community project under GPL, blkreplay is open to |
| 74 |
contributions from hardware vendors, other data centers, |
| 75 |
the kernel hacker community, etc. |
| 76 |
|
| 77 |
------------------------------------------------------- |
| 78 |
|
| 79 |
Full documentation: |
| 80 |
https://github.com/schoebel/blkreplay/raw/master/doc/blkreplay.pdf |
| 81 |
|
| 82 |
Sourcecode: https://github.com/schoebel/blkreplay/ |
| 83 |
|
| 84 |
Database of loads: http://www.blkreplay.org/loads/ |
| 85 |
|