23.07.2014 Views

Lustre 1.6 Operations Manual

Lustre 1.6 Operations Manual

Lustre 1.6 Operations Manual

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

22.3.19 Drawbacks in Doing Multi-client O_APPEND<br />

Writes<br />

It is possible to do multi-client O_APPEND writes to a single file, but there are few<br />

drawbacks that may make this a sub-optimal solution. These drawbacks are:<br />

■<br />

■<br />

■<br />

Each client needs to take an EOF lock on all the OSTs, as it is difficult to know<br />

which OST holds the end of the file until you check all the OSTs. As all the clients<br />

are using the same O_APPEND, there is significant locking overhead.<br />

The second client cannot get all locks until the end of the writing of the first<br />

client, as the taking serializes all writes from the clients.<br />

To avoid deadlocks, the taking of these locks occurs in a known, consistent order.<br />

As a client cannot know which OST holds the next piece of the file until the client<br />

has locks on all OSTS, there is a need of these locks in case of a striped file.<br />

22.3.20 Slowdown Occurs During <strong>Lustre</strong> Startup<br />

When <strong>Lustre</strong> starts, the <strong>Lustre</strong> filesystem needs to read in data from the disk. For the<br />

very first mdsrate run after the reboot, the MDS needs to wait on all the OSTs for<br />

object precreation. This causes a slowdown to occur when <strong>Lustre</strong> starts up.<br />

After the filesystem has been running for some time, it contains more data in cache<br />

and hence, the variability caused by reading critical metadata from disk is mostly<br />

eliminated. The filesystem now reads data from the cache.<br />

22.3.21 Log Message ‘Out of Memory’ on OST<br />

When planning the hardware for an OSS node, consider the memory usage of<br />

several components in the <strong>Lustre</strong> system. If insufficient memory is available, an ‘out<br />

of memory’ message can be logged.<br />

During normal operation, several conditions indicate insufficient RAM on a server<br />

node:<br />

■ kernel "Out of memory" and/or "oom-killer" messages<br />

■<br />

■<br />

<strong>Lustre</strong> "kmalloc of 'mmm' (NNNN bytes) failed..." messages<br />

<strong>Lustre</strong> or kernel stack traces showing processes stuck in "try_to_free_pages"<br />

For information on determining the MDS’s memory and OSS memory requirements,<br />

see Memory Requirements.<br />

Chapter 22 <strong>Lustre</strong> Troubleshooting Tips 22-17

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!