EMAIL EFFICIENCY PROBLEM
Email architectures rely heavily on the performance of servers and disks. Over time, the email has become a very demanding application because of the size of emails and more group collaboration over email. The data requirements of email applications are projected to grow from KBs 10 years ago to MBs now, and to GBs in the future. Without any control over data management, email applications are not benefiting from the advances in high performance storage.
Studies indicate that efficient retrieval of recent data matters a lot to users. Similarly, stale data can be served with higher latency. Searching and indexing requirements for email have also grown in the last decade. In a large workgroup collaboration, sharing of documents and emails that go out to a large number of users with large attachments can stress the servers.
Datagres technology eases the data management issues in the email infrastructure. The added benefits have a clear and visible impact on the user's expectation from the email environment. At the same time, the servers can now host more user-load hence lowering the cost of providing email.
The case study in this document demonstrates the value proposition for Zimbra Email servers. However, it can be used with other email servers as well.
ZIMBRA EMAIL CASE STUDY
Accelerating the performance of Zimbra Email Server
Zimbra email server offers multiple opportunities for system acceleration. Zimbra receives messages, and writes it to the disk. Zimbra then transfers these messages into the message-store and MySQL database. Some deployments also provide the Lucene indexes. Zimbra authenticates on the LDAP server. Authentication is mostly reads but sometimes the last-login-time etc. may also be stored.
Similarly while reading information from Zimbra via WEBMAIL, POP or IMAP protocols, the app server performs authentication on the LDAP and reads the messages from the MySQL server and the message-store on the file system. The same process is followed for downloading attachments.
User behavior in the email services follows the pattern of Last-In-First-Read, most of the time. The most recent data is required to be read first as confirmed by the behavior of the WEBMAIL, POP and SMTP protocols.
TEST TO CONFIRM IF ACCELERATION IS POSSIBLE INSIDE ZIMBRA ARCHITECTURE
The results below show that the overall gain in both SMTP and POP transactions in 'Ops per minute' is 3X with a very low cost Intel SATA SSD and PerfAccel Software from Datagres that intelligently manages the data for acceleration purposes.
This means 3X faster WEBMAIL, POP and SMTP transactions on the same server architecture. Not only does the end-user observe faster email but the server utilization is also reduced by 3 times. Datagres PerfAccel thus provides an overall amplification of 9X times for Zimbra deployment. No integration with Zimbra is required from the Datagres PerfAccel Software perspective.
The total-ops is the sum of all transactions that happen inside Zimbra, given a POP or SMTP workload:
OVERALL PERFORMANCE GAIN - ZIMBRA WITH DATAGRES PERFACCEL
With PCIe SSD cards it is also possible to accelerate SMTP and the load of incoming email to the server, to a much higher acceleration factor.
|Following are the key features of the Datagres PerfAccel Software|
|Intelligent caching accelerates data access by reducing the time taken for an IO operation|
|The customer's email files are intelligently pre-fetched in the cache at 9am before the rush builds up|
|When a large email attachment is sent to a large number of users, the concurrent reads for that email return much faster|
|More number of concurrent users are hosted per server and the cost of the server is reduced|
|Index files are kept in the cache forever|
|Logging servers use write-back caching|
|Number of queries on MySQL can be increased to thousands per second|
|Multi Tenant Control relieves user from locking SSDs only for one application or for one device|
|Relieves users from the burden of manual management of data on SSDs|
|Can accelerate currently existing data on ext2, ext3 or NFS file systems|
|Supports iSCSI devices|
|To understand the caching behavior, customers can use analytical mode to understand the application workload without any SSD|
|Caching mechanism is completely manageable by the user|
|File Level Cache Management is available|
|Single Point of Management Interface|
|Graphical Usage Analysis available to trace IO bottlenecks|