Darpan,
You're pretty lucky - I was writing a performance test to judge the latency of
our memcached implementation so that we'd know what to expect when its rolled
out to production. The total time represents the time it takes to perform the
operation on all N elements, while the mean time is the average time of a
single operation. For our tests, N=20,000 is good enough to get an accurate
idea.
Comments appreciated.
Best regards,
Ben
* Using standard Java client.
* All scenarios are executed concurrently (only 2 presently).
* A random set of keys are used (removes cache locality issues).
* Using application code path to simulate real-world requests (local caches
invalidated).
** Environment #1 is running an older build than environment #2.
#1) Performance environment (client: 4x-Xeon 3.2GHz, 4GB, Red Hat 3.4.6;
server: 4x-Xeon 2.8GHz, 4GB, Red Hat 3.4.6, Oracle 9i, caches: 2x-1GB instances
on equivalent machine as client)
A) Compression enabled.
---------- Scenario: Customer Test ----------
- Put:
* Remote Cache (total): 0:03:37.647 (h:m:s.mi)
* Remote Cache (mean): 10,882,350 (ns)
- Retrieval:
* Database (total): 0:00:27.029 (h:m:s.mi)
* Database (mean): 1,351,450 (ns)
* Remote Cache (total): 0:00:37.900 (h:m:s.mi)
* Remote Cache (mean): 1,895,000 (ns)
- Removal:
* Remote Cache (total): 0:00:23.912 (h:m:s.mi)
* Remote Cache (mean): 1,195,600 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,880
* Mean element size: 10.4326171875Kb
---------- Scenario: Customer View Test ----------
- Put:
* Remote Cache (total): 0:05:24.534 (h:m:s.mi)
* Remote Cache (mean): 0:00:00.016 (h:m:s.mi)
- Retrieval:
* Database (total): 0:00:26.767 (h:m:s.mi)
* Database (mean): 1,338,350 (ns)
* Remote Cache (total): 0:00:41.900 (h:m:s.mi)
* Remote Cache (mean): 2,095,000 (ns)
- Removal:
* Remote Cache (total): 0:00:24.681 (h:m:s.mi)
* Remote Cache (mean): 1,234,050 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,880
* Mean element size: 203.923828125Kb
B) Compression disabled.
---------- Scenario: Customer Test ----------
- Put:
* Remote Cache (total): 0:01:52.540 (h:m:s.mi)
* Remote Cache (mean): 5,627,000 (ns)
- Retrieval:
* Database (total): 0:00:26.721 (h:m:s.mi)
* Database (mean): 1,336,050 (ns)
* Remote Cache (total): 0:00:31.198 (h:m:s.mi)
* Remote Cache (mean): 1,559,900 (ns)
- Removal:
* Remote Cache (total): 0:00:22.722 (h:m:s.mi)
* Remote Cache (mean): 1,136,100 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,880
* Mean element size: 11.0693359375Kb
---------- Scenario: Customer View Test ----------
- Put:
* Remote Cache (total): 0:02:11.640 (h:m:s.mi)
* Remote Cache (mean): 6,582,000 (ns)
- Retrieval:
* Database (total): 0:00:26.283 (h:m:s.mi)
* Database (mean): 1,314,150 (ns)
* Remote Cache (total): 0:00:30.462 (h:m:s.mi)
* Remote Cache (mean): 1,523,100 (ns)
- Removal:
* Remote Cache (total): 0:00:22.828 (h:m:s.mi)
* Remote Cache (mean): 1,141,400 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,880
* Mean element size: 203.8330078125Kb
#2) Development machine (4x-Xeon 2.33GHz, 4GB, Red Hat 3.4.6, Oracle 9i)
A) Compression enabled.
---------- Scenario: Customer Test ----------
- Put:
* Remote Cache (total): 0:02:37.791 (h:m:s.mi)
* Remote Cache (mean): 7,889,550 (ns)
- Retrieval:
* Single element request:
- Database (total): 0:02:47.563 (h:m:s.mi)
- Database (mean): 8,378,150 (ns)
- Remote Cache (total): 0:01:58.538 (h:m:s.mi)
- Remote Cache (mean): 5,926,900 (ns)
* Bulk element request:
- Remote Cache (total): 0:00:22.731 (h:m:s.mi)
- Removal:
* Remote Cache (total): 0:00:06.154 (h:m:s.mi)
* Remote Cache (mean): 307,700 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,885
* Mean element size: 10.3037109375Kb
---------- Scenario: Customer View Test ----------
- Put:
* Remote Cache (total): 0:04:16.113 (h:m:s.mi)
* Remote Cache (mean): 0:00:00.012 (h:m:s.mi)
- Retrieval:
* Single element request:
- Database (total): 0:04:33.528 (h:m:s.mi)
- Database (mean): 0:00:00.013 (h:m:s.mi)
- Remote Cache (total): 0:02:38.933 (h:m:s.mi)
- Remote Cache (mean): 7,946,650 (ns)
* Bulk element request:
- Remote Cache (total): 0:00:09.849 (h:m:s.mi)
- Removal:
* Remote Cache (total): 0:00:04.977 (h:m:s.mi)
* Remote Cache (mean): 248,850 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,885
* Mean element size: 203.1474609375Kb
B) Compression disabled.
---------- Scenario: Customer Test ----------
- Put:
* Remote Cache (total): 0:01:15.580 (h:m:s.mi)
* Remote Cache (mean): 3,779,000 (ns)
- Retrieval:
* Single element request:
- Database (total): 0:02:29.240 (h:m:s.mi)
- Database (mean): 7,462,000 (ns)
- Remote Cache (total): 0:01:39.165 (h:m:s.mi)
- Remote Cache (mean): 4,958,250 (ns)
* Bulk element request:
- Remote Cache (total): 0:00:33.387 (h:m:s.mi)
- Removal:
* Remote Cache (total): 0:00:09.171 (h:m:s.mi)
* Remote Cache (mean): 458,550 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,885
* Mean element size: 10.1787109375Kb
---------- Scenario: Customer View Test ----------
- Put:
* Remote Cache (total): 0:01:42.935 (h:m:s.mi)
* Remote Cache (mean): 5,146,750 (ns)
- Retrieval:
* Single element request:
- Database (total): 0:04:22.415 (h:m:s.mi)
- Database (mean): 0:00:00.013 (h:m:s.mi)
- Remote Cache (total): 0:02:14.100 (h:m:s.mi)
- Remote Cache (mean): 6,705,000 (ns)
* Bulk element request:
- Remote Cache (total): 0:00:50.595 (h:m:s.mi)
- Removal:
* Remote Cache (total): 0:00:04.712 (h:m:s.mi)
* Remote Cache (mean): 235,600 (ns)
- Test Statistics:
* Number of elements used: 20,000
* Number of available elements: 96,885
* Mean element size: 204.166015625Kb
----- Original Message ----
From: Darpan Dinker <[EMAIL PROTECTED]>
To: [email protected]
Sent: Saturday, July 21, 2007 1:02:54 AM
Subject: memcacheD performance
Can somebody provide a ballpark response time for a get( ) under light
load for:
- a memcached server on the same system as the client
- a remove server with 1 Gigabit Ethernet network
I am guessing that it is on the order of 20ms and 150ms.
Feedback appreciated; please include the processor, memory size and OS.
Darpan
--
Blog: http://darpanetwork.blogspot.com/
Atom: http://darpanetwork.blogspot.com/feeds/posts/default
____________________________________________________________________________________
Shape Yahoo! in your own image. Join our Network Research Panel today!
http://surveylink.yahoo.com/gmrs/yahoo_panel_invite.asp?a=7
