Shulou(Shulou.com)05/31 Report--

This article mainly introduces what Ceph needs for computer hardware, the article is very detailed, has a certain reference value, interested friends must read it!

Introduction to the requirements of Ceph for computer hardware

I read the official documentation of Ceph, which mentions Ceph's need for hardware resources, as well as in Learning Ceph. Let me make a brief summary here. Ceph mainly consists of the following components:

MON

OSD

MDS

RGW

The focus of the discussion is also on the degree to which components are dependent on CPU and memory.

MON MON is responsible for managing CRUSHMAP, which is not CPU-intensive and relies heavily on memory because of the need for fast access to data. CPU can be configured with low memory, but with sufficient memory, 1GB memory for each process instance.

OSD OSD runs the rados service, calculates the location of objects through the CRUSH algorithm, and requires a reasonable number of CPU to manage data and its copies. OSD data is not particularly dependent on memory, and normally each process instance is 500MB. In abnormal cases (data migration recovery, etc.), each process instance, each 1TB data corresponds to 1GB memory. So the more memory, the better. My OSD process is running normally.

MDS metadata server, which serves CephFS, is a CPU-intensive process that takes up a lot of CPU resources, so it is necessary to equip MDS with CPU with strong performance. Memory requirements are also large because of the need for fast access to data; 1GB memory for each process instance.

There are also some factors that affect storage performance: a large number of read and write requests will affect performance for a single disk drive; different file systems will also affect system performance; let's talk about the selection of disk drives.

In the process of using the system, I found that the memory usage of various components: osd: about 2G, coincidence will go up to 7G; mon: at present, around 700m, it once soared to 8G, which also led to a shortage of system memory. Later, I restarted the mon process one by one; rgw: 20GB of memory currently occupied

Mechanical disk:

The recommended disk size is as small as 1TB

The larger the disk, the smaller the storage space per unit of GB, but the more memory is required, especially in recovery,backfill and reblance

It is not recommended to divide a disk into multiple regions and run multiple OSD processes

OSD and mon are not recommended, or OSD and mds are running on the same hard drive.

Many problems with slow osd are often due to the overuse of disk devices, so please use dedicated disks to install the operating system and software; dedicated disk devices for osd daemon and dedicated disk devices for journal.

SSD disk: the use of SSD can effectively improve system performance, but before deciding to use SSD, you need to consider several issues: for the case of high write performance, to buy a suitable SSD, some cheap SSD, may not be as good as a good mechanical hard disk Sequential writing: when you use a piece of SSD as the journal of multiple disks, it is possible for several OSD to write journal at the same time, while writing journal is sequential writing, so you need to pay attention to sequential write performance; pay attention to partition alignment

Nic: at least 2 1Gbps NICs, one for public network to interact with users of the system, and one for cluster network for copying data copies, recovery, backfill, etc.

These are all the contents of the article "what are the requirements of Ceph for computer hardware?" Thank you for your reading! Hope to share the content to help you, more related knowledge, welcome to follow the industry information channel!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.