Background
Assuming that you have already selected your Fredhopper deployment model, this document serves as a guideline to determine the sizing of your hardware for on-premise deployments. The Fredhopper deployment model consists of online servers that provide responses to user queries, and an offline server for processing and indexing data. The offline server also allows you to configure Fredhopper. This document does not cover virtualization environments.
This document applies to Fredhopper 6.1 and higher.
Requirements
You will need the following information to size the hardware:
- Pageviews in the peak month
- If the number of items is below or above 100,000 items
- If you require a concurrent or cascading deployment model (see decision guide)
Also, verify that the assumption is correct that 65% of your page views will require you to query Fredhopper.
Considerations
The table below lists the key considerations that impact Fredhopper hardware sizing:
| Considerations | |
|---|---|
| Number of products | We distinguish between set-ups of <100k and of >100k products, including product variations |
| Deployment model | The cascading deployment model uses double indexers in the off-line server and therefore requires 4 GB more RAM - see step 3. |
| Fail-safeness | All recommendations are based on the assumption that you run Fredhopper behind a load balancer - as Fredhopper is stateless - in a fail-safe mode: even if one server fails the system will still respond - although at a lower capacity. |
| Incremental updates capacity | The capacity to process incremental updates varies between deployment models and the number of products in the set-up - see step 3. |
| Number of universes | The number of universes in Fredhopper does not impact the hardware sizing |
| Number of Locales | The number of Locales in Fredhopper reduces the capacity, e.g. 100,000 items in a single Locale require the same hardware as 50,000 items in 2 Locales or 10,000 items in 10 Locales (assuming non-shared products across the Locales). |
Step 1: Chose the number of servers
Please select the number of on-line and off-line servers from the table below, based on the number of page views:
| Page views in peak month | Fredhopper QPS | < 100k products (on-line + off-line) |
>100k products* (on-line + off-line) |
|---|---|---|---|
| Up to 10,000,000 | 10 | 2** + 1 = 3 | 2** + 1 = 3 |
| Up to 20,000,000 | 20 | 2** + 1 = 3 | 2** + 1 = 3 |
| Up to 30,000,000 | 30 | 2** + 1 = 3 | 2 + 1 = 3 |
| Up to 40,000,000 | 40 | 2** + 1 = 3 | 2 + 1 = 3 |
| Up to 50,000,000 | 50 | 2** + 1 = 3 | 3 + 1 = 4 |
| Up to 60,000,000 | 60 | 2 + 1 = 3 | 3 + 1 = 4 |
| Up to 80,000,000 | 80 | 2 + 1 = 3 | 4 + 1 = 5 |
| Up to 100,000,000 | 100 | 2 + 1 = 3 | 5 + 1 = 6 |
| Up to 120,000,000 | 120 | 3 + 1 = 4 | 6 + 1 = 7 |
** We have added one extra server to create a fail-safe set-up - see table in Considerations.
 |
Fredhopper supports per machine 50 Queries per Second (QPS) for <100.000 products and 20 QPS for >100.000 products. QPS in the table above is based on Peak Page Views (Peak PV) with the following calculation: QPS = Peak PV/month * 5% (peak day) * 11% (peak hour) * 65% (average Fredhopper page share). Note: Specific set-ups may require different assumptions. |
Step 2: Select your hardware configuration
Fredhopper uses commodity hardware. Please select the required hardware configuration based on the table below:
| <100k products | >100k products | |
|---|---|---|
| CPU* | 2x Dual / 1x Quad Core AMD Opteron 2.2+ GHz or 2x Dual / 1x Quad Core Intel Xeon Processor 3.0+ GHz |
2x Quad-Core AMD Opteron 2.6+ GHz or 2x Quad Core Intel Xeon Processor 3.0+ GHz |
| Disk** | On-line: 70 GB effective space Off-line: 250 GB effective space |
On-line 70 GB effective space Off-line 250 GB effective space |
| Network cards | 1Gbit | 1Gbit |
| Sample system | Dell PowerEdge SC1435 HP ProLiant DL360 G5 2x dual core |
Dell PowerEdge 2950 HP ProLiant DL360 G5 2x quad core (459960-005) |
** It is recommended to use a SATA 7,200 RPM disk. For set-ups with >1 million products, a SCSI (SAS) 15,000 RPM disk is recommended for faster disk access. For shorter downtime in case of any disk crashes or no downtime at all, you may consider using RAID-1 disks on the indexer machine.
|
Fredhopper's Querying SLA is a response time of <200ms for 95% of all queries, and <500ms for 99.9% of all queries - in standard test conditions. Response times may vary due to the type of queries and data. |
Step 3: Select your RAM
Please select the required RAM for your on-line and off-line machines, based on the table below:
| Deployment model | <100k products | >100k products* |
|---|---|---|
| Concurrent | On-line 4 GB Off-line 8 GB |
On-line 8 GB Off-line 8 GB |
| Cascading | On-line 4 GB Off-line 12 GB |
On-line 8 GB Off-line 12 GB |
The table below lists the capacity to process incremental updates depending on the deployment model and the number of products:
| Deployment model | <100k products | >100k products |
|---|---|---|
| Concurrent | 250 per second | 500 per second |
| Cascading | 125 per second | 250 per second |
Kommentare
0 Kommentare
Bitte melden Sie sich an, um einen Kommentar zu hinterlassen.