In my MIX talk I told that for 2k reads, we take 3ms. It is actually the case under load where the servers are straining. Here is a latest performance shot that we got with the CTP3 bits.
Bulk APIs: Note that these numbers are without any bulk operations. With bulk gets (we are trying to add a Bulk Get in RTM timeframe within a region) we get upwards of 211,000 operations per second. (assuming a 32 key per batch).
Local Cache: None of these tests use Local Cache = true. That will again give upwards of 200K operations per second.
[Statutory Warning: These are the numbers that we got in our test labs to give you an idea of the performance characteristics of this product. There is no implicit warranty or guarantee on these numbers!]
Servers : 2
Server Configuration :
Processor : AMD Phenom (9600 B) Quad Core Processor
RAM size : 8 GB RAM
Operating System : 64 bit win 2k3
Network : 1GBPS network
Secondary : 0
Client configuration : Routing with NO local cache.
No of Agents : 10
Throughput :
Scenario
object
Number of Clients
RC4(Ops/sec)
CTP2(Ops/sec)
Mainly Gets
Byte[] (2kb)
100
34500
36300
Byte[](2kb to 16kb)
150
27000
27300
Put
Char[](2kb)
200
9170
6060
Put-Get
String[](1kb)
11500
8000
Put-Get-Remove
Byte[](4kb) wrapped in object
30300
29880
Latency :
Operation(byte array)
Object
RC4(In ms)
CTP2(in ms)
Add
Char [](2kb)
1.35
1.59
Byte [](2kb)
0.63
0.55
Get
0.50
0.51
Remove
0.49
0.44
Trending with and without secondary throughput on RC4:
with secondary
without secondary
Put-Get(2k)
String[](1k)
4300
Put-Get-Remove(2k)
8280
--
Murali