VMware CLOUD DIRECTOR 1.0 Manuel d'utilisateur télécharger pdf (Page 23)

PERFORMANCE STUDY /23

VMware vCloud Director 1.0 Performance and Best Practices

For deploying without fence mode, the throughput is higher than with fence mode because less configuration

needs to occur. Here, too, throughput keeps increasing when the concurrency level increases.

Performance Tuning Notes

The latency to deploy a vApp in fence mode does not increase much when there are more VMs in the vApp. This

is because fence mode adds a fixed cost to the deployment time (deploy and configure a vShield Edge VM one

time) and the increase in time is not proportional to the number of VMs in the vApp.

Deploying multiple vApps concurrently can achieve high throughput.

Inventory Sync

When a vCloud Director cell is initialized, the current vCenter Server inventory information is retrieved and the

relevant data is stored in the empty vCloud Director database. This process is called initial sync. The vCloud

Director server may be shut down and restarted. When it is restarted, all the current vCenter Server inventory

information is retrieved again. Changes to the inventory data are stored in the vCloud Director Database. This

process is called restart-cell-sync. The difference between initial sync and restart-cell-sync is the state of the

vCloud Director database. Initial sync starts with an empty vCloud Director database, whereas restart-cell-sync

starts with a database that already contains the inventory information. After restart-cell-sync, the vCloud Director

database is periodically updated.

The vCenter Server process may also be shut down and restarted. When this occurs, the vCloud Director server

tries to reconnect to the vCenter Server and re-sync the inventory information. This process is called reconnect-

vCenter-sync. Reconnect-vCenter-sync takes less time than restart-cell-sync because, in reconnect-vCenter-

sync, the vCloud Director server has a cache that stores the inventory information in memory.

Sync Latency for Different Inventory Sizes

Figure 21 shows that the latencies for initial sync and restart-cell-sync increase linearly as the number of virtual

machines in the system grow. For reconnect-vCenter-sync latency, the default inventory cache is not big enough

to hold 3000 inventory objects, so some inventory objects must be fetched from the vCloud Director database.

This process causes the significant latency increase seen between 2000 and 3000 inventory sizes.

For the restart-cell-sync experiment, there are no changes for the inventory objects, and so there are no

database inserts required. In contrast, the latency for initial sync is higher because inventory objects need to be

inserted into the vCloud Director database. The latency for initial sync can be longer than the latency for restart-

cell-sync even when the inventory size is relatively small. The majority of time is spent comparing the difference

between the stored inventory object in the vCloud Director database with the one in vCenter Server.

For reconnect-vCenter-sync, because the in-memory inventory cache can potentially hold all or most of the

inventory objects, no extra time is needed to fetch these objects from the vCloud Director database. This is why

reconnect-vCenter-sync gives the best performance among the three types of inventory syncs. The following

sections describe more about the inventory cache effects and how to tune the inventory cache for better

performance.

1 2 ... 18 19 20 21 22 23 24 25 26 27 28 ... 31 32

Pas de commentaire

VMware CLOUD DIRECTOR 1.0 Manuel d'utilisateur Page 23