Re: Nutch segment merge is very slow
I never merge segments on my single box, too slow worse always getting into HD full. Means you have to ditch older segments after while though... So you need to keep segments at least younger than maxcrawl duration 2010/4/6, arkadi.kosmy...@csiro.au arkadi.kosmy...@csiro.au: Hi, -Original Message- From: Susam Pal [mailto:susam@gmail.com] Sent: Tuesday, 6 April 2010 12:18 AM To: nutch-user@lucene.apache.org Subject: Re: Nutch segment merge is very slow On Mon, Apr 5, 2010 at 5:27 PM, ashokkumar.raveendi...@wipro.com wrote: Hi I'm using Nutch crawler in my project and crawled more than 2GB of data using Nutch runbot script. Up to 2GB segment merger has took and ended with in 24 hrs but now it takes more than 48 hrs and still running. I have set depth to 16 and topN to 2500. I want to run crawler every day as per my requirement. How to speed up segment merge and index process. Regards Ashokkumar.R Hi, From my experience of running Nutch on a single box to crawl a corporate intranet with a depth as high as 16 and a topN value greater than 1000, I feel it isn't feasible to have one crawl per day. That is, if you consider your site a monolithic object and try to recrawl the whole site each time. Normally, web sites are not homogeneous. To keep your index up to date, you only have to recrawl regularly the parts that change fast, which is 1% to 10% of the site, and refresh other parts perhaps once in every few months. One of these options might help you. 1. Run Nutch on a Hadoop cluster to distribute the job and speed up processing. Then you have to run your web server on a cluster as well because it will have to serve all contents of your site every day + serve other clients. 2. Reduce the crawl depth to about 7, 8, or whatever works for you. This means you wouldn't be crawling links discovered in the crawl perform at depth 8. This may be a good or a bad thing for you depending on whether you want to crawl URLs found so deep in the crawl. These URLs may be obscure and less important because they are so many hops away from your seed URLs. Loosing quality. Can it be a good thing? ` 3. However, if the URLs found very deep are also important and you want to crawl them, you might have to sacrifice low ranking URLs by setting a smaller topN value, say, 1000, or whatever works for you. Loosing quality in a different way. How do you calculate ranks? Link based methods do not work nearly as well on intranets as on the global Web. Regards, Susam Pal Regards, Arkadi Kosmynin CSIRO Astronomy and Space Science -- -MilleBii-
Re: Nutch segment merge is very slow
On Mon, Apr 5, 2010 at 5:27 PM, ashokkumar.raveendi...@wipro.com wrote: Hi I'm using Nutch crawler in my project and crawled more than 2GB of data using Nutch runbot script. Up to 2GB segment merger has took and ended with in 24 hrs but now it takes more than 48 hrs and still running. I have set depth to 16 and topN to 2500. I want to run crawler every day as per my requirement. How to speed up segment merge and index process. Regards Ashokkumar.R Hi, From my experience of running Nutch on a single box to crawl a corporate intranet with a depth as high as 16 and a topN value greater than 1000, I feel it isn't feasible to have one crawl per day. One of these options might help you. 1. Run Nutch on a Hadoop cluster to distribute the job and speed up processing. 2. Reduce the crawl depth to about 7, 8, or whatever works for you. This means you wouldn't be crawling links discovered in the crawl perform at depth 8. This may be a good or a bad thing for you depending on whether you want to crawl URLs found so deep in the crawl. These URLs may be obscure and less important because they are so many hops away from your seed URLs. ` 3. However, if the URLs found very deep are also important and you want to crawl them, you might have to sacrifice low ranking URLs by setting a smaller topN value, say, 1000, or whatever works for you. Regards, Susam Pal
RE: Nutch segment merge is very slow
Hi, Thank you for your suggestion. I have around 500+ internet urls configured for crawling and crawl process is running in Amazon cloud. I have already reduced my depth to 8, topN to 1000 and also increased fetcher threads to 150 and limited 50 urls per host using generate.max.per.host property. With this configuration Generate, Fetch, Parse, Update completes in max 10 hrs. When comes to segment merge it takes lot of time. As a temporary solution I am not doing the segment merge and directly indexing the fetched segments. With this solution I am able to finish the crawl process with in 24hrs. Now I am looking for long term solution to optimize segment merge process. Regards Ashokkumar.R -Original Message- From: Susam Pal [mailto:susam@gmail.com] Sent: Monday, April 05, 2010 7:48 PM To: nutch-user@lucene.apache.org Subject: Re: Nutch segment merge is very slow On Mon, Apr 5, 2010 at 5:27 PM, ashokkumar.raveendi...@wipro.com wrote: Hi I'm using Nutch crawler in my project and crawled more than 2GB of data using Nutch runbot script. Up to 2GB segment merger has took and ended with in 24 hrs but now it takes more than 48 hrs and still running. I have set depth to 16 and topN to 2500. I want to run crawler every day as per my requirement. How to speed up segment merge and index process. Regards Ashokkumar.R Hi, From my experience of running Nutch on a single box to crawl a corporate intranet with a depth as high as 16 and a topN value greater than 1000, I feel it isn't feasible to have one crawl per day. One of these options might help you. 1. Run Nutch on a Hadoop cluster to distribute the job and speed up processing. 2. Reduce the crawl depth to about 7, 8, or whatever works for you. This means you wouldn't be crawling links discovered in the crawl perform at depth 8. This may be a good or a bad thing for you depending on whether you want to crawl URLs found so deep in the crawl. These URLs may be obscure and less important because they are so many hops away from your seed URLs. ` 3. However, if the URLs found very deep are also important and you want to crawl them, you might have to sacrifice low ranking URLs by setting a smaller topN value, say, 1000, or whatever works for you. Regards, Susam Pal Please do not print this email unless it is absolutely necessary. The information contained in this electronic message and any attachments to this message are intended for the exclusive use of the addressee(s) and may contain proprietary, confidential or privileged information. If you are not the intended recipient, you should not disseminate, distribute or copy this e-mail. Please notify the sender immediately and destroy all copies of this message and any attachments. WARNING: Computer viruses can be transmitted via email. The recipient should check this email and any attachments for the presence of viruses. The company accepts no liability for any damage caused by any virus transmitted by this email. www.wipro.com
Re: Nutch segment merge is very slow
On 2010-04-05 16:54, ashokkumar.raveendi...@wipro.com wrote: Hi, Thank you for your suggestion. I have around 500+ internet urls configured for crawling and crawl process is running in Amazon cloud. I have already reduced my depth to 8, topN to 1000 and also increased fetcher threads to 150 and limited 50 urls per host using generate.max.per.host property. With this configuration Generate, Fetch, Parse, Update completes in max 10 hrs. When comes to segment merge it takes lot of time. As a temporary solution I am not doing the segment merge and directly indexing the fetched segments. With this solution I am able to finish the crawl process with in 24hrs. Now I am looking for long term solution to optimize segment merge process. Segment merging is not strictly necessary, unless you have a hundred segments or so. If this step takes too much time, but still the number of segments is well below a hundred, just don't merge them. -- Best regards, Andrzej Bialecki ___. ___ ___ ___ _ _ __ [__ || __|__/|__||\/| Information Retrieval, Semantic Web ___|||__|| \| || | Embedded Unix, System Integration http://www.sigram.com Contact: info at sigram dot com
RE: Nutch segment merge is very slow
Hi, -Original Message- From: Susam Pal [mailto:susam@gmail.com] Sent: Tuesday, 6 April 2010 12:18 AM To: nutch-user@lucene.apache.org Subject: Re: Nutch segment merge is very slow On Mon, Apr 5, 2010 at 5:27 PM, ashokkumar.raveendi...@wipro.com wrote: Hi I'm using Nutch crawler in my project and crawled more than 2GB of data using Nutch runbot script. Up to 2GB segment merger has took and ended with in 24 hrs but now it takes more than 48 hrs and still running. I have set depth to 16 and topN to 2500. I want to run crawler every day as per my requirement. How to speed up segment merge and index process. Regards Ashokkumar.R Hi, From my experience of running Nutch on a single box to crawl a corporate intranet with a depth as high as 16 and a topN value greater than 1000, I feel it isn't feasible to have one crawl per day. That is, if you consider your site a monolithic object and try to recrawl the whole site each time. Normally, web sites are not homogeneous. To keep your index up to date, you only have to recrawl regularly the parts that change fast, which is 1% to 10% of the site, and refresh other parts perhaps once in every few months. One of these options might help you. 1. Run Nutch on a Hadoop cluster to distribute the job and speed up processing. Then you have to run your web server on a cluster as well because it will have to serve all contents of your site every day + serve other clients. 2. Reduce the crawl depth to about 7, 8, or whatever works for you. This means you wouldn't be crawling links discovered in the crawl perform at depth 8. This may be a good or a bad thing for you depending on whether you want to crawl URLs found so deep in the crawl. These URLs may be obscure and less important because they are so many hops away from your seed URLs. Loosing quality. Can it be a good thing? ` 3. However, if the URLs found very deep are also important and you want to crawl them, you might have to sacrifice low ranking URLs by setting a smaller topN value, say, 1000, or whatever works for you. Loosing quality in a different way. How do you calculate ranks? Link based methods do not work nearly as well on intranets as on the global Web. Regards, Susam Pal Regards, Arkadi Kosmynin CSIRO Astronomy and Space Science